Library
Condensed editions and reading notes with links to the original papers.
691 papers
Newest added first-
BWM-Forge: Agentic World Programming for Embodied Intelligence
Preprint posted · arXiv (preprint; technical report)
World models · 14 min text · 42 citations
-
Temporal Straightening for Latent Planning
· International Conference on Machine Learning (ICML 2026)
World models · 10 min text · 77 citations
-
SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotors
Preprint posted · arXiv (preprint; under review)
World models · 8 min text · 64 citations
-
AdaJEPA: An Adaptive Latent World Model
Preprint posted · arXiv (preprint)
World models · 7 min text · 46 citations
-
Self-supervised DXA representations encode multi-system disease risk, biological aging and heritability
Preprint posted · arXiv (preprint)
Scientific machine learning · 9 min text · 73 citations
-
LpWM: A Case for Sparse Representations in World Models
Preprint posted · arXiv (preprint)
World models · 7 min text · 46 citations
-
LeHami: Symplectic Latent Dynamics for World Models
Preprint posted · arXiv (preprint)
World models · 9 min text · 37 citations
-
LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics
Preprint posted · arXiv (preprint)
World models · 9 min text · 33 citations
-
HP-JEPA: Hierarchical Partitioning for Multi-Resolution Graph Joint-Embedding Predictive Learning
Preprint posted · arXiv (preprint)
Graph Learning · 5 min text · 26 citations
-
Music-JEPA: Learning a World Model of Sound from Action
Preprint posted · arXiv (preprint)
Speech and audio · 4 min text · 42 citations
-
Crys-JEPA: Accelerating Crystal Discovery via Embedding Screening and Generative Refinement
Preprint posted · arXiv (preprint)
Scientific machine learning · 6 min text · 60 citations
-
AD-E2E-JEPA: A Joint-Embedding Predictive Architecture For End-to-End Autonomous Driving
· arXiv (preprint)
World models · 5 min text · 61 citations
-
Preserving Unstable Modes Through Inverse Dynamics in JEPA World Models
· arXiv (preprint)
World models · 7 min text · 41 citations
-
H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning
· arXiv (preprint)
World models · 11 min text · 93 citations
-
Causal-JEPA: Learning World Models through Object-Level Latent Masking
Preprint posted · ICML 2026 (accepted)
World models · 12 min text · 52 citations
-
Parallel Stochastic Gradient-Based Planning for World Models
· arXiv (preprint)
World models · 7 min text · 42 citations
-
Value-guided action planning with JEPA world models
· arXiv (preprint)
World models · 5 min text · 18 citations
-
What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?
· Transactions on Machine Learning Research
World models · 12 min text · 99 citations
-
A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures
Preprint posted · ICLR 2026 World Models Workshop (accepted)
Representation learning · 5 min text · 55 citations
-
Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models
· Neural Information Processing Systems (NeurIPS 2025)
World models · 11 min text · 83 citations
-
DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning
· International Conference on Machine Learning (ICML 2025)
World models · 8 min text · 64 citations
-
Semantic Tube Prediction: Beating LLM Data Efficiency with JEPA
Preprint posted · arXiv (preprint)
Language models · 7 min text · 56 citations
-
S-JEPA : Soft Clustering Anchors for Self-Supervised Speech Representation Learning
Preprint posted · arXiv (preprint)
Speech and audio · 6 min text · 42 citations
-
Soft Clustering Anchors for Self-Supervised Speech Representation Learning in Joint Embedding Prediction Architectures
Preprint posted · arXiv (preprint)
Speech and audio · 6 min text · 32 citations
-
JEPA as a Neural Tokenizer: Learning Robust Speech Representations with Density Adaptive Attention
Preprint posted · UniReps workshop at NeurIPS 2025
Speech and audio · 7 min text · 19 citations
-
VL-JEPA: Joint Embedding Predictive Architecture for Vision-language
Preprint posted · ICLR 2026
Vision-language models · 5 min text · 85 citations
-
LLM-JEPA: Large Language Models Meet Joint Embedding Predictive Architectures
Preprint posted · ICLR 2026
Language models · 7 min text · 53 citations
-
Back to the Features: DINO as a Foundation for Video World Models
Preprint posted · arXiv (preprint)
World models · 9 min text · 69 citations
-
Intuitive physics understanding emerges from self-supervised pretraining on natural videos
Preprint posted · arXiv (preprint)
Representation learning · 5 min text · 64 citations
-
Video Representation Learning with Joint-Embedding Predictive Architectures
Preprint posted · arXiv (preprint)
Representation learning · 8 min text · 44 citations
-
Learning and Leveraging World Models in Visual Representation Learning
Preprint posted · arXiv (preprint)
Representation learning · 7 min text · 58 citations
-
When Does LeJEPA Learn a World Model?
Preprint posted · NeurIPS 2026 (accepted)
Learning theory · 12 min text · 103 citations
-
Rectified LpJEPA: Joint-Embedding Predictive Architectures with Sparse and Maximum-Entropy Representations
Preprint posted · ICML 2026 (accepted)
Learning theory · 10 min text · 75 citations
-
Gaussian Embeddings: How JEPAs Secretly Learn Your Data Density
Preprint posted · arXiv (preprint)
Learning theory · 3 min text · 17 citations
-
Joint Embedding Predictive Architectures Focus on Slow Features
Preprint posted · NeurIPS 2022 SSL Theory and Practice Workshop
Representation learning · 6 min text · 40 citations
-
Introduction to Latent Variable Energy-Based Models: A Path Towards Autonomous Machine Intelligence
· Journal of Statistical Mechanics: Theory and Experiment
World models · 9 min text · 62 citations
-
Hierarchical Planning with Latent World Models
Preprint posted · CoRL 2026 (accepted)
World models · 6 min text · 52 citations
-
LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels
· arXiv (preprint)
World models · 8 min text · 51 citations
-
LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
Preprint posted · arXiv (preprint)
Learning theory · 10 min text · 121 citations
-
V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning
Preprint posted · arXiv (preprint)
World models · 12 min text · 139 citations
-
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
Preprint posted · arXiv (preprint)
World models · 13 min text · 132 citations
-
Revisiting Feature Prediction for Learning Visual Representations from Video
Preprint posted · arXiv (preprint)
World models · 8 min text · 81 citations
-
MC-JEPA: A Joint-Embedding Predictive Architecture for Self-Supervised Learning of Motion and Content Features
Preprint posted · arXiv (preprint)
World models · 5 min text · 87 citations
-
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
· CVPR 2023
World models · 7 min text · 79 citations
-
A Path Towards Autonomous Machine Intelligence
· Position paper (OpenReview)
World models · 10 min text · 105 citations
-
Training Verifiers to Solve Math Word Problems
Preprint posted · arXiv (preprint)
Language models · 7 min text · 29 citations
-
Measuring Massive Multitask Language Understanding
Preprint posted · arXiv (preprint)
Language models · 7 min text · 32 citations
-
Proximal Policy Optimization Algorithms
Preprint posted · arXiv (preprint)
Language models · 6 min text · 14 citations
-
DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
Preprint posted · arXiv (preprint)
Language models · 11 min text · 61 citations
-
Direct Preference Optimization: Your Language Model is Secretly a Reward Model
Preprint posted · NeurIPS 2023
Language models · 10 min text · 51 citations
-
Flow Matching for Generative Modeling
· ICLR 2023
Generative modeling · 11 min text · 52 citations
-
Training Compute-Optimal Large Language Models
· arXiv preprint; subsequently NeurIPS 2022
Scaling laws · 12 min text · 60 citations
-
Predicting trends in the quality of state-of-the-art neural networks without access to training or testing data
· Nature Communications 12, 4122
Model diagnostics · 13 min text · 43 citations
-
Traditional and Heavy-Tailed Self Regularization in Neural Network Models
· ICML 2019 · PMLR 97
Learning theory and diagnostics · 8 min text · 50 citations
-
Muon is Scalable for LLM Training
· Technical report
Machine learning optimization · 11 min text · 45 citations
-
Decoupled Weight Decay Regularization
· ICLR 2019
Machine learning optimization · 7 min text · 28 citations
-
Shampoo: Preconditioned Stochastic Tensor Optimization
· ICML 2018, Proceedings of Machine Learning Research 80:1842–1850
Optimization · 9 min text · 24 citations
-
YaRN: Efficient Context Window Extension of Large Language Models
Preprint first posted · ICLR 2024
Language models · 9 min text · 40 citations
-
Extending Context Window of Large Language Models via Position Interpolation
Preprint first posted · arXiv
Language models · 7 min text · 43 citations
-
RoFormer: Enhanced Transformer with Rotary Position Embedding
Preprint first posted · arXiv
Language models · 8 min text · 43 citations
-
mHC: Manifold-Constrained Hyper-Connections
Conference began · ICML 2026 · PMLR 306
Language models · 7 min text · 44 citations
-
Hopfield Networks is All You Need
Preprint first posted · ICLR 2021
Language models · 20 min text · 119 citations
-
Concerning nonnegative matrices and doubly stochastic matrices
Published · Pacific Journal of Mathematics 21(2), 343–348
Deep Learning Foundations · 4 min text · 6 citations
-
Natural Gradient Works Efficiently in Learning
Published · Neural Computation 10(2), 251–276
Deep Learning Foundations · 7 min text · 41 citations
-
Robust Uncertainty Principles: Exact Signal Reconstruction from Highly Incomplete Frequency Information
· IEEE Transactions on Information Theory 52(2 (489–509
Deep Learning Foundations · 11 min text · 23 citations
-
Compressed Sensing
· IEEE Transactions on Information Theory 52(4), 1289–1306
Deep Learning Foundations · 12 min text · 47 citations
-
Extensions of Lipschitz Mappings into a Hilbert Space
Published · Contemporary Mathematics 26, 189–206
Deep Learning Foundations · 6 min text · 14 citations
-
The Bitter Lesson
Published · Author’s website (research essay)
Deep Learning Foundations · 1 min text · 0 citations
-
Embodied First Exam
Preprint posted · arXiv (preprint)
AI agents · 35 min text · 203 citations
-
Explicitly capturing demonstrator multi-modality for distillation into diffusion policies
Preprint posted · CoRL 2026 Pretrain to Adapt Workshop
Robotics · 11 min text · 34 citations
-
ToolDex: An Ego-Exo Human Demonstration Dataset for Creative Physical Tool Use
Preprint posted · arXiv (preprint)
Robotics · 13 min text · 68 citations
-
BEHAVIOR-Spatial Benchmark: Evaluating Spatial Understanding for Robotics
Preprint posted · arXiv (preprint)
Robotics · 15 min text · 43 citations
-
Text-to-Image Generative Modeling Without Language Pretraining
Preprint posted · arXiv (preprint)
Computer Vision · 9 min text · 41 citations
-
SteerAct: Unlocking Action Models through Few-Shot Temporal Steering
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 58 citations
-
Dream What I Can: Sparse Visual Planning for Long-Horizon Robot Manipulation
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 35 citations
-
A Language-Informed Loss for Visual Generation
Preprint posted · arXiv (preprint)
Computer Vision · 12 min text · 54 citations
-
TRACER: A Multimodal Harness for Patient Care in Hospitals
Preprint posted · arXiv (preprint)
Healthcare AI · 16 min text · 40 citations
-
T2-z: Test-Time Optimization of Environment Latents for Robotic World Models
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 49 citations
-
Generative Representation Learning for Scalable Latent Tokenization
Preprint posted · arXiv (preprint)
Computer Vision · 10 min text · 62 citations
-
I3L: Iterative Interactive Imitation Learning for Whole-Body Mobile Manipulation
Preprint posted · arXiv (preprint)
Robotics · 18 min text · 62 citations
-
FACT: A Multimodal Force-Action-Track Foundation Model for Human-to-Robot Dexterity
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 38 citations
-
RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 45 citations
-
Cross-Embodiment Robot Foundation World Models with Latent Actions
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 44 citations
-
OpenWAM: An Open Framework for Composable World-Action Models
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 108 citations
-
LoGo: Local-Global Rewards for Consistent Long-Horizon Video Generation
Preprint posted · arXiv (preprint)
Computer Vision · 11 min text · 45 citations
-
Wrench-ACT: Enhancing Robot Policies for Contact Rich Behavior Using Direct Wrench Control
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 38 citations
-
T²Mem: Learning Test-Time Memory for Robotics
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 21 citations
-
DexAgent: An Agentic Human2Sim2Robot Framework for Dexterous Manipulation with Self-Evolving Tool Library
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 37 citations
-
TACTIC: Understanding Tactile Encoders and Conditioning for Contact-rich Robot Manipulation Policies
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 37 citations
-
What do VLM-Based Vision-Language Navigation Models Rely on: Interpreting and Steering Policy Behavior
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 37 citations
-
Norm2Tex: Augmenting Visuo-Tactile Simulations with Texture
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 31 citations
-
Adaptive World Memory 3D Foundation Model for Scalable 3D Mapping, Localization, and Rendering
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 82 citations
-
Planning along Differentiable Charts of Constraint Manifolds with General-Purpose IK Solvers
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 49 citations
-
DUET-DINO: Simultaneous Cross-View World Modeling for Latent Planning in Robot Manipulation
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 33 citations
-
One Demonstration, Many Objects: Generalizing Manipulation via Local Contact Geometry
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 49 citations
-
TrAct: Bridging Robot Control and Visual Prediction with Visual Tracks
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 43 citations
-
Revisiting Open-Loop Execution in Robotics: Toward Reactive, Higher-Performing Policies
Preprint posted · arXiv (preprint)
Robotics · 13 min text · 75 citations
-
Zero-Shot Mission-Level Evaluation for Aerial MLLM Agents
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 33 citations
-
Masked Visual Actions for Unified World Modeling
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 82 citations
-
MapDreamer: Aerial Imagery Conditioned Latent Diffusion for Lane-Level Map Generation
Preprint posted · ECCV 2026 (accepted)
Robotics · 12 min text · 31 citations
-
SimFoundry: Modular and Automated Scene Generation for Policy Learning and Evaluation
Preprint posted · arXiv (preprint)
Robotics · 17 min text · 106 citations
-
Training and Evaluating Diffusion Policies with Long Context Lengths
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 38 citations
-
T-Rex: Tactile-Reactive Dexterous Manipulation
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 73 citations
-
Semidefinite Relaxations for Collision-Free Motion Planning
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 51 citations
-
DuoBench: A Reproducible Benchmark for Bimanual Manipulation in Simulation and the Real World
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 49 citations
-
Ambient Diffusion Policy: Imitation Learning from Suboptimal Data in Robotics
Preprint posted · arXiv (preprint)
Robotics · 18 min text · 94 citations
-
Shape Formation for the Cooperative Transportation of Arbitrary Objects Using Multi-Agent Reinforcement Learning
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 22 citations
-
SAGrid: Scaling Robot Simulation through Automatic Affordance Annotation on In-the-Wild 3D Assets
· ICRA 2026
Robotics · 7 min text · 27 citations
-
Robot Control Stack: A Lean Ecosystem for Robot Learning at Scale
· ICRA 2026
Robotics · 9 min text · 31 citations
-
MetricNet: Recovering Metric Scale in Generative Navigation Policies
· ICRA 2026
Robotics · 9 min text · 46 citations
-
IMPASTO: Integrating Model-Based Planning with Learned Dynamics Models for Robotic Oil Painting Reproduction
· ICRA 2026
Robotics · 11 min text · 53 citations
-
How Well do Diffusion Policies Learn Kinematic Constraint Manifolds?
· ICRA 2026
Robotics · 7 min text · 40 citations
-
Dream2Flow: Bridging Video Generation and Open-World Manipulation with 3D Object Flow
· ICRA 2026
Robotics · 12 min text · 93 citations
-
Planning with the Views
Preprint posted · arXiv (preprint)
Robotics · 18 min text · 86 citations
-
GPIC: A Giant Permissive Image Corpus for Visual Generation
Preprint posted · arXiv (preprint)
Computer Vision · 7 min text · 37 citations
-
Faster Algorithms for Growing Collision-Free Convex Polytopes in Robot Configuration Space
· The International Journal of Robotics Research
Robotics · 7 min text · 28 citations
-
GCS*: Forward Heuristic Search on Implicit Graphs of Convex Sets
· WAFR 2024 proceedings (2026 edition)
Robotics · 12 min text · 27 citations
-
ESI-Bench: Towards Embodied Spatial Intelligence that Closes the Perception-Action Loop
Preprint posted · arXiv (preprint)
Robotics · 20 min text · 32 citations
-
StereoPolicy: Improving Robotic Manipulation Policies via Stereo Perception
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 78 citations
-
A framework of digital biomarkers for neurodegenerative diseases
· Nature Reviews Bioengineering
Healthcare AI · 16 min text · 216 citations
-
HumanScore: Benchmarking Human Motions in Generated Videos
Preprint posted · NeurIPS 2026, Evaluations and Datasets
Computer Vision · 16 min text · 81 citations
-
Robotik
· Handbuch der Künstlichen Intelligenz (book chapter)
Robotics · Reading note
-
A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation
· Science Robotics
Robotics · 16 min text · 86 citations
-
RAGEN-2: Reasoning Collapse in Agentic RL
Preprint posted · arXiv (preprint)
AI agents · 15 min text · 74 citations
-
Augmented Reality for RObots (ARRO): Pointing Visuomotor Policies Towards Visual Robustness
· IEEE Robotics and Automation Letters
Robotics · 8 min text · 53 citations
-
MIRAGE: The Illusion of Visual Understanding
Preprint posted · arXiv (preprint)
Healthcare AI · 12 min text · 54 citations
-
MARCUS: An agentic, multimodal vision-language model for cardiac diagnosis and management
Preprint posted · arXiv (preprint)
Healthcare AI · 16 min text · 39 citations
-
CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation
Preprint posted · arXiv (preprint)
AI agents · 16 min text · 103 citations
-
Rewarding DINO: Predicting Dense Rewards with Vision Foundation Models
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 38 citations
-
OGScene3D: Incremental Open-Vocabulary 3D Gaussian Scene Graph Mapping for Scene Understanding
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 85 citations
-
Enabling Dynamic Tracking in Vision-Language-Action Models via Time-Discrete and Time-Continuous Velocity Feedforward
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 30 citations
-
EchoAtlas: A Conversational, Multi-View Vision-Language Foundation Model for Echocardiography Interpretation and Clinical Reasoning
Preprint posted · medRxiv (preprint)
Healthcare AI · 12 min text · 34 citations
-
FlowTouch: View-Invariant Visuo-Tactile Prediction
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 39 citations
-
Cost-Driven Representation Learning for Linear Quadratic Gaussian Control: Part II
Preprint posted · arXiv (preprint)
Control theory · 10 min text · 38 citations
-
Goal Achievement Guided Exploitation: A Principled Performance-Based Scheduling Framework for Reinforcement Learning
· Transactions on Machine Learning Research
Robotics · Reading note
-
Learning from Trials and Errors: Reflective Test-Time Planning for Embodied LLMs
Preprint posted · arXiv (preprint)
AI agents · 17 min text · 46 citations
-
Latent Forcing: Reordering the Diffusion Trajectory for Pixel-Space Image Generation
Preprint posted · arXiv (preprint)
Computer Vision · 8 min text · 57 citations
-
SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes
Preprint posted · ICML 2026 (accepted)
Robotics · 24 min text · 46 citations
-
A Framework for Combining Optimization-Based and Analytic Inverse Kinematics
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 96 citations
-
VLAgents: A Policy Server for Efficient VLA Inference
Preprint posted · arXiv (preprint)
Robotics · 2 min text · 16 citations
-
VideoWeave: A Data-Centric Approach for Efficient Video Understanding
Preprint posted · arXiv (preprint)
Computer Vision · 4 min text · 49 citations
-
VLIC: Vision-Language Models As Perceptual Judges for Human-Aligned Image Compression
· CVPR 2026
Computer Vision · 10 min text · 57 citations
-
Theory of Space: Can Foundation Models Construct Spatial Beliefs through Active Exploration?
· ICLR 2026
Robotics · 12 min text · 67 citations
-
Recreating Video Arenas via Automated Preference Scoring
· NeurIPS 2026 Evaluations and Datasets (accepted)
Computer Vision · 8 min text · 42 citations
-
ROSETTA: Constructing Code-Based Reward from Unconstrained Language Preference
· ICLR 2026
Robotics · 16 min text · 56 citations
-
QuantiPhy: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models
· CVPR 2026
Computer Vision · 15 min text · 60 citations
-
PointWorld: Scaling 3D World Models for In-The-Wild Robotic Manipulation
· CVPR 2026
Robotics · 15 min text · 175 citations
-
MindCube: Spatial Mental Modeling from Limited Views
· ICLR 2026
Computer Vision · 18 min text · 77 citations
-
Cambrian-S: Towards Spatial Supersensing in Video
· ICLR 2026
Robotics · 18 min text · 168 citations
-
Large Video Planner Enables Generalizable Robot Control
Preprint posted · arXiv (preprint)
Robotics · 13 min text · 91 citations
-
EchoGraph system for automated quality assessment of echocardiography reports
· npj Digital Medicine
Healthcare AI · 7 min text · 24 citations
-
Polytope Inner Approximations with Fewer Faces Subject to Minimum-Volume, Point-Inclusion, and Set-Intersection Constraints
· CDC 2025
Robotics · Reading note
-
Label-Efficient Point Cloud Segmentation with Active Learning
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 30 citations
-
ENACT: Evaluating Embodied Cognition with World Modeling of Egocentric Interaction
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 88 citations
-
MRLGraph: A Novel, Graph-Based Map Relation Learning Network for Traffic Element to Lane Assignment Based on Geometric Map Data
· ITSC 2025
Robotics · Reading note
-
Multi-Query Shortest-Path Problem in Graphs of Convex Sets
· WAFR 2024 proceedings (2026 print edition)
Robotics · 9 min text · 37 citations
-
WorldScore: A Unified Evaluation Benchmark for World Generation
· ICCV 2025
Computer Vision · 13 min text · 98 citations
-
Scalable Real2Sim: Physics-Aware Asset Generation Via Robotic Pick-and-Place Setups
· IROS 2025
Robotics · 8 min text · 39 citations
-
Sampling-Based Motion Planning with Discrete Configuration-Space Symmetries
· IROS 2025
Robotics · 10 min text · 54 citations
-
Repurposing 2D Diffusion Models with Gaussian Atlas for 3D Generation
· ICCV 2025
Computer Vision · 11 min text · 61 citations
-
Refined Policy Distillation: From VLA Generalists to RL Experts
· IROS 2025
Robotics · 6 min text · 33 citations
-
REGRACE: A Robust and Efficient Graph-based Re-localization Algorithm using Consistency Evaluation
· IROS 2025
Robotics · 9 min text · 36 citations
-
Label-Efficient LiDAR Panoptic Segmentation
· IROS 2025
Robotics · 4 min text · 44 citations
-
FlowNav: Combining Flow Matching and Depth Priors for Efficient Navigation
· IROS 2025
Robotics · 7 min text · 51 citations
-
Flow to the Mode: Mode-Seeking Diffusion Autoencoders for State-of-the-Art Image Tokenization
· ICCV 2025
Computer Vision · 9 min text · 64 citations
-
Empirical Analysis of Sim-and-Real Cotraining of Diffusion Policies for Planar Pushing from Pixels
· IROS 2025
Robotics · 8 min text · 47 citations
-
ConPoSe: LLM-Guided Contact Point Selection for Scalable Cooperative Object Pushing
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 21 citations
-
EEG-Based Brain-Computer Interface for Robotic Assistance with User Intention Prediction
Preprint posted · Research Square (preprint)
Robotics · 9 min text · 27 citations
-
Articulated Object Estimation in the Wild
Preprint posted · CoRL 2025 (accepted)
Robotics · 7 min text · 60 citations
-
"Data will solve robotics and automation: True or false?": A debate
· Science Robotics
Robotics · 4 min text · 5 citations
-
DiWA: Diffusion Policy Adaptation with World Models
Preprint posted · CoRL 2025 (accepted)
Robotics · 9 min text · 73 citations
-
Advancing Science- and Evidence-based AI Policy
· Science
AI policy · 4 min text · 15 citations
-
A roadmap for AI in robotics
Published · Nature Machine Intelligence 7, 818–824
Robotics · 6 min text · 47 citations
-
Mixed Discrete and Continuous Planning using Shortest Walks in Graphs of Convex Sets
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 63 citations
-
The California Report on Frontier AI Policy
Preprint posted · arXiv (preprint)
AI policy · 14 min text · 145 citations
-
An Updated Tutorial on Graph-Based SLAM
· Robotics Goes MOOC: Knowledge (book chapter)
Robotics · Reading note
-
Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces
· CVPR 2025
Computer Vision · 13 min text · 107 citations
-
The Language of Motion: Unifying Verbal and Non-verbal Language of 3D Human Motion
· CVPR 2025
Computer Vision · 7 min text · 87 citations
-
T*: Re-thinking Temporal Search for Long-Form Video Understanding
· CVPR 2025
Computer Vision · 15 min text · 93 citations
-
Multimodal Spatial Language Maps for Robot Navigation and Manipulation
Preprint posted · The International Journal of Robotics Research (accepted)
Robotics · 9 min text · 99 citations
-
Foundation versus domain-specific models for left ventricular segmentation on cardiac ultrasound
· npj Digital Medicine
Healthcare AI · 6 min text · 44 citations
-
Planning Shorter Paths in Graphs of Convex Sets by Undistorting Parametrized Configuration Spaces
· IEEE Robotics and Automation Letters
Robotics · 9 min text · 33 citations
-
VLM-Vac: Enhancing Smart Vacuums through VLM Knowledge Distillation and Language-Guided Experience Replay
· ICRA 2025
Robotics · 5 min text · 31 citations
-
UAD: Unsupervised Affordance Distillation for Generalization in Robotic Manipulation
· ICRA 2025
Robotics · 10 min text · 132 citations
-
Proximity and Visuotactile Point Cloud Fusion for Contact Patches in Extreme Deformation
· ICRA 2025
Robotics · 7 min text · 34 citations
-
LUMOS: Language-Conditioned Imitation Learning with World Models
· ICRA 2025
Robotics · 10 min text · 37 citations
-
CloudTrack: Scalable UAV Tracking with Cloud Semantics
· ICRA 2025
Robotics · 6 min text · 28 citations
-
Chain-of-Modality: Learning Manipulation Programs from Multimodal Human Videos with Vision-Language-Models
· ICRA 2025
Robotics · 6 min text · 68 citations
-
ARCap: Collecting High-quality Human Demonstrations for Robot Learning with Augmented Reality Feedback
· ICRA 2025
Robotics · 4 min text · 49 citations
-
A Real-to-Sim-to-Real Approach to Robotic Manipulation with VLM-Generated Iterative Keypoint Rewards
· ICRA 2025
Robotics · 8 min text · 101 citations
-
A New Semidefinite Relaxation for Linear and Piecewise-Affine Optimal Control with Time Scaling
· ICRA 2025
Robotics · 6 min text · 49 citations
-
Steerable Scene Generation with Post Training and Inference-Time Search
Preprint posted · arXiv (preprint)
Robotics · 15 min text · 56 citations
-
Dexterous Contact-Rich Manipulation via the Contact Trust Region
Preprint posted · arXiv (preprint)
Robotics · 17 min text · 90 citations
-
RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning
Preprint posted · arXiv (preprint)
AI agents · 14 min text · 73 citations
-
Altered gait stability and regularity in lower limb amputees observed across different body segments
· Journal of Biomechanics
Biomechanics · 9 min text · 55 citations
-
Superfast Configuration-Space Convex Set Computation on GPUs for Online Motion Planning
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 54 citations
-
BYE: Build Your Encoder with One Sequence of Exploration Data for Long-Term Dynamic Scene Understanding
· IEEE Robotics and Automation Letters
Robotics · 7 min text · 50 citations
-
Evaluating large language models in echocardiography reporting: opportunities and challenges
· European Heart Journal – Digital Health
Healthcare AI · 8 min text · 48 citations
-
NeuHMR: Neural Rendering-Guided Human Motion Reconstruction
· 3DV 2025
Computer Vision · 9 min text · 56 citations
-
iPack: Intuitive Bin Packing with Large Language Models
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 29 citations
-
Towards Fine-Grained Video Question Answering
Preprint posted · arXiv (preprint)
Computer Vision · 9 min text · 60 citations
-
BEHAVIOR Robot Suite: Streamlining Real-World Whole-Body Manipulation for Everyday Household Activities
Preprint posted · CoRL 2025 (accepted)
Robotics · 15 min text · 177 citations
-
Physics-Driven Data Generation for Contact-Rich Manipulation via Trajectory Optimization
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 89 citations
-
LiDAR Registration with Visual Foundation Models
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 33 citations
-
Developing ICU Clinical Behavioral Atlas Using Ambient Intelligence and Computer Vision
· NEJM AI
Healthcare AI · 8 min text · 25 citations
-
A Good Foundation is Worth Many Labels: Label-Efficient Panoptic Segmentation
· IEEE Robotics and Automation Letters
Robotics · 7 min text · 37 citations
-
s1: Simple test-time scaling
· EMNLP 2025
Language models · 14 min text · 90 citations
-
Rendering Humans Behind Occlusions
· IEEE Transactions on Pattern Analysis and Machine Intelligence
Computer Vision · 8 min text · 67 citations
-
VAGEN: Reinforcing World Model Reasoning for Multi-Turn VLM Agents
· NeurIPS 2025
AI agents · 19 min text · 82 citations
-
Should VLMs be Pre-trained with Image Data?
· ICLR 2025
Computer Vision · 11 min text · 53 citations
-
Non-Euclidean Motion Planning with Graphs of Geodesically-Convex Sets
· The International Journal of Robotics Research
Robotics · 12 min text · 51 citations
-
History-Guided Video Diffusion
· ICML 2025
Computer Vision · 17 min text · 70 citations
-
Exploring Diffusion Transformer Designs via Grafting
· NeurIPS 2025
Computer Vision · 12 min text · 52 citations
-
End-to-end 2D-3D Registration between Image and LiDAR Point Cloud for Vehicle Localization
· IEEE Transactions on Robotics
Robotics · 9 min text · 65 citations
-
Discovering Latent Graphs with GFlowNets for Diverse Conditional Image Generation
· NeurIPS 2025
Computer Vision · 12 min text · 87 citations
-
An Interactive Agent Foundation Model
· CVPR Workshops 2025
AI agents · 8 min text · 64 citations
-
LetsMap: Unsupervised Representation Learning for Label-Efficient Semantic BEV Mapping
· ECCV 2024 proceedings (2025 print edition)
Robotics · 9 min text · 38 citations
-
Photorealistic Video Generation with Diffusion Models
· ECCV 2024 proceedings (2025 print edition)
Computer Vision · 7 min text · 89 citations
-
Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
· The International Journal of Robotics Research
Robotics · 12 min text · 70 citations
-
ForceBand: Learning Forceful Manipulation with sEMG
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 46 citations
-
Guava: An Effective and Universal Harness for Embodied Manipulation
Preprint posted · arXiv (preprint)
AI agents · 7 min text · 35 citations
-
Discovering Hybrid World Representations with Co-Evolving Foundation Models
· AAAI 2026
World models · 4 min text · 40 citations
-
SIMPACT: Simulation-Enabled Action Planning using Vision-Language Models
· CVPR 2026
Robotics · 9 min text · 73 citations
-
Composition-Grounded Data Synthesis for Visual Reasoning
· ICLR 2026
Multimodal reasoning · 10 min text · 76 citations
-
V-GRPO: Online Reinforcement Learning for Denoising Generative Models Is Easier than You Think
Preprint posted · CVPR 2026 Findings (accepted)
Efficient Learning · 13 min text · 47 citations
-
ActionEQA: Action Interface for Embodied Question Answering
· TMLR 2026
AI agents · 20 min text · 82 citations
-
Multi-Modal Manipulation via Multi-Modal Policy Consensus
· ICRA 2026
Robotics · 9 min text · 43 citations
-
Adversarial Game-Theoretic Algorithm for Dexterous Grasp Synthesis
· ICRA 2026
Robotics · 9 min text · 21 citations
-
Learning, Reasoning, and Planning with Neuro-Symbolic Concepts
Submitted · PhD dissertation, MIT
Reasoning · 98 min text · 550 citations
-
Implicit State Estimation via Video Replanning
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 30 citations
-
Learning Linear Attention in Polynomial Time
· NeurIPS 2025
Learning theory · 23 min text · 47 citations
-
Creativity or Brute Force? Using Brainteasers as a Window into the Problem-Solving Abilities of Large Language Models
· NeurIPS 2025
Language models · 18 min text · 41 citations
-
Finding structure in logographic writing with library learning II: Grapheme, sound, and meaning systematicity
· CogSci 2025
Scientific discovery · 7 min text · 33 citations
-
Agent Workflow Memory
· ICML 2025
AI agents · 13 min text · 31 citations
-
Generating Fine Details of Entity Interactions
· EMNLP 2025 Industry Track
Computer Vision · 11 min text · 35 citations
-
One-Shot Manipulation Strategy Learning by Making Contact Analogies
· ICRA 2025
Robotics · 6 min text · 65 citations
-
KALM: Keypoint Abstraction Using Large Models for Object-Relative Imitation Learning
· ICRA 2025
Robotics · 8 min text · 56 citations
-
Infer Human's Intentions Before Following Natural Language Instructions
· AAAI 2025
AI agents · 11 min text · 35 citations
-
What Makes a Maze Look Like a Maze?
· ICLR 2025
Multimodal reasoning · 9 min text · 40 citations
-
Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making
· NeurIPS 2024 Datasets and Benchmarks Track
AI agents · 25 min text · 101 citations
-
Hybrid Declarative-Imperative Representations for Hybrid Discrete-Continuous Decision-Making
· WAFR 2024 proceedings
Robotics · 12 min text · 28 citations
-
Finding structure in logographic writing with library learning
· CogSci 2024
Scientific discovery · 6 min text · 49 citations
-
Learning Iterative Reasoning through Energy Diffusion
· ICML 2024
Reasoning · 7 min text · 55 citations
-
“Set It Up!”: Functional Object Arrangement with Compositional Generative Models
· International Journal of Robotics Research (IJRR) 2025
Robotics · 18 min text · 45 citations
-
Learning Adaptive Planning Representations with Natural Language Guidance
· ICLR 2024
Robotics · 7 min text · 50 citations
-
Learning to Act from Actionless Videos through Dense Correspondences
· ICLR 2024
Robotics · 11 min text · 51 citations
-
Learning Planning Abstractions from Language
· ICLR 2024
Robotics · 8 min text · 62 citations
-
What Planning Problems Can A Relational Neural Network Solve?
· NeurIPS 2023
Reasoning · 15 min text · 32 citations
-
What's Left? Concept Grounding with Logic-Enhanced Foundation Models
· NeurIPS 2023
Reasoning · 7 min text · 48 citations
-
Learning Reusable Manipulation Strategies
· CoRL 2023
Robotics · 7 min text · 35 citations
-
Compositional Diffusion-Based Continuous Constraint Solvers
· CoRL 2023
Robotics · 10 min text · 37 citations
-
Composable Part-Based Manipulation
· CoRL 2023
Robotics · 9 min text · 61 citations
-
NS3D: Neuro-Symbolic Grounding of 3D Objects and Relations
· CVPR 2023
Reasoning · 8 min text · 40 citations
-
Learning Rational Subgoals from Demonstrations and Instructions
· AAAI 2023
AI agents · 10 min text · 47 citations
-
DisCo: Improving Compositional Generalization in Visual Reasoning through Distribution Coverage
· TMLR 2023
Multimodal reasoning · 9 min text · 40 citations
-
On the Expressiveness and Generalization of Hypergraph Neural Networks
· LoG 2022 (extended abstract)
Graph Learning · 10 min text · 35 citations
-
Sparse and Local Networks for Hypergraph Reasoning
· LoG 2022
Graph Learning · 8 min text · 44 citations
-
PDSketch: Integrated Domain Programming, Learning, and Planning
· NeurIPS 2022
Robotics · 8 min text · 50 citations
-
HandMeThat: Human-Robot Communication in Physical and Social Environments
· NeurIPS 2022 Datasets and Benchmarks Track
AI agents · 10 min text · 50 citations
-
CLEVRER-Humans: Describing Physical and Causal Events the Human Way
· NeurIPS 2022 Datasets and Benchmarks Track
Multimodal reasoning · 9 min text · 59 citations
-
IKEA-Manual: Seeing Shape Assembly Step by Step
· NeurIPS 2022 Datasets and Benchmarks Track
Computer Vision · 11 min text · 59 citations
-
Translating a Visual LEGO Manual to a Machine-Executable Plan
· ECCV 2022
Computer Vision · 10 min text · 52 citations
-
Programmatic Concept Learning for Human Motion Description and Synthesis
· CVPR 2022
Computer Vision · 7 min text · 64 citations
-
FALCON: Fast Visual Concept Learning by Integrating Images, Linguistic Descriptions, and Conceptual Relations
· ICLR 2022
Reasoning · 12 min text · 45 citations
-
Grammar-Based Grounded Lexicon Learning
· NeurIPS 2021
Multimodal reasoning · 13 min text · 51 citations
-
Temporal and Object Quantification Networks
· IJCAI 2021
Reasoning · 9 min text · 34 citations
-
Language-Mediated, Object-Centric Representation Learning
· ACL 2021 (Findings)
Multimodal reasoning · 8 min text · 47 citations
-
Hierarchical Motion Understanding via Motion Programs
· CVPR 2021
Computer Vision · 8 min text · 34 citations
-
Grounding Physical Concepts of Objects and Events Through Dynamic Visual Reasoning
· ICLR 2021
Multimodal reasoning · 11 min text · 55 citations
-
Object-Centric Diagnosis of Visual Reasoning
Preprint posted · arXiv (preprint)
Multimodal reasoning · 8 min text · 68 citations
-
Multi-Plane Program Induction with 3D Box Priors
· NeurIPS 2020
Computer Vision · 9 min text · 48 citations
-
Perspective Plane Program Induction From a Single Image
· CVPR 2020
Computer Vision · 6 min text · 60 citations
-
Visual Concept-Metaconcept Learning
· NeurIPS 2019
Reasoning · 9 min text · 52 citations
-
Program-Guided Image Manipulators
· ICCV 2019
Computer Vision · 8 min text · 57 citations
-
Visually Grounded Neural Syntax Acquisition
· ACL 2019
Multimodal reasoning · 8 min text · 79 citations
-
Neurally-Guided Structure Inference
· ICML 2019
Reasoning · 7 min text · 40 citations
-
Unified Visual-Semantic Embeddings: Bridging Vision and Language With Structured Meaning Representations
· CVPR 2019
Multimodal reasoning · 8 min text · 54 citations
-
The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervision
· ICLR 2019
Reasoning · 12 min text · 52 citations
-
Neural Logic Machines
· ICLR 2019
Reasoning · 11 min text · 75 citations
-
Neural Phrase-to-Phrase Machine Translation
Preprint posted · arXiv (preprint)
Language models · 6 min text · 36 citations
-
Acquisition of Localization Confidence for Accurate Object Detection
· ECCV 2018
Computer Vision · 5 min text · 33 citations
-
Learning Visually-Grounded Semantics from Contrastive Adversarial Samples
· COLING 2018
Multimodal reasoning · 6 min text · 40 citations
-
Universal Agent for Disentangling Environments and Tasks
· ICLR 2018
Robotics · Reading note
-
What Can Help Pedestrian Detection?
· CVPR 2017
Computer Vision · 6 min text · 35 citations
-
Designing Accessible Robot Communication for Blind People
Workshop · InterAI Workshop at CHI 2026
Robotics · Reading note
-
Advancement of concept maturity level for a servicing architecture for the Habitable Worlds Observatory
· SPIE Optics + Photonics
Robotics · Reading note
-
CRATER: A Regolith Collection Robot for In-Situ Manufacturing of Lightweight Metamaterials
· AIAA Regional Student Conferences
Robotics · Reading note
-
The Second Half of Autonomous Driving: Debating the Road to N-Nines Reliability
· TechRxiv (preprint)
Robotics · Reading note
-
Global reconstruction of three decades of fine-grained nighttime light data with analysis of large-scale infrastructure and landmarks
· Remote Sensing of Environment
Remote sensing · Reading note
-
Evolution path of multimodal collaborative optimization and training techniques for end-to-end autonomous driving systems
Published online · Scientia Sinica Technologica 55(10), 1638–1658
Robotics · 9 min text · 66 citations
-
Position: Prospective of Autonomous Driving - Multimodal LLMs, World Models, Embodied Intelligence, AI Alignment, and Mamba
· WACV (Workshops) 2025
Robotics · 8 min text · 207 citations
-
Understanding and Imitating Human-Robot Motion with Restricted Visual Fields
· IROS 2025
Robotics · 6 min text · 32 citations
-
MultiNash-PF: A Particle Filtering Approach for Computing Multiple Local Generalized Nash Equilibria in Trajectory Games
· IROS 2025
Robotics · 5 min text · 39 citations
-
Multi-Agent Inverse Reinforcement Learning in Real World Unstructured Pedestrian Crowds
· IROS 2025
Robotics · 7 min text · 84 citations
-
Active Inverse Learning in Stackelberg Trajectory Games
· ACC 2025
Robotics · 6 min text · 47 citations
-
Robots Learning Through Physical Interactive Intelligence
· AAAI 2025
Robotics · 1 min text · 9 citations
-
Principles and Guidelines for Evaluating Social Robot Navigation Algorithms
· ACM Transactions on Human-Robot Interaction
Robotics · 19 min text · 168 citations
-
SARO: Space-Aware Robot System for Terrain Crossing via Vision-Language Model
· ICRA 2025
Robotics · 9 min text · 68 citations
-
Robust Robot Walker: Learning Agile Locomotion over Tiny Traps
· ICRA 2025
Robotics · 9 min text · 43 citations
-
Generalizing Motion Planners with Mixture of Experts for Autonomous Driving
· ICRA 2025
Robotics · 5 min text · 38 citations
-
Playful DoggyBot: Learning Agile and Precise Quadrupedal Locomotion
· IROS 2025
Robotics · 7 min text · 55 citations
-
SPACE: Speaker Adaptation for Acoustic Eavesdropping Using mmWave Radio Signals
· IEEE Transactions on Mobile Computing
Audio and acoustics · 10 min text · 77 citations
-
Gen-NCAP: A Generative Simulator for Corner Case Benchmarking in End-to-End Autonomous Driving
· IASEAI 2026
Robotics · 8 min text · 41 citations
-
Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes
· Annual Review of Control, Robotics, and Autonomous Systems
Robotics · 17 min text · 212 citations
-
BUMBLE: Unifying Reasoning and Acting with Vision-Language Models for Building-wide Mobile Manipulation
· ICRA 2025
Robotics · 8 min text · 78 citations
-
FLaRe: Achieving Masterful and Adaptive Robot Policies with Large-Scale Reinforcement Learning Fine-Tuning
· ICRA 2025
Robotics · 8 min text · 68 citations
-
CurricuLLM: Automatic Task Curricula Design for Learning Complex Robot Skills Using Large Language Models
· ICRA 2025
Robotics · 6 min text · 55 citations
-
Risk-Sensitive Orbital Debris Collision Avoidance using Distributionally Robust Chance Constraints
· AIAA SciTech Forum 2025
Robotics · 9 min text · 40 citations
-
When Should a Leader Act Suboptimally? The Role of Inferability in Repeated Stackelberg Games
· IEEE Transactions on Automatic Control
Robotics · 14 min text · 34 citations
-
Strategic Decision-Making in Multiagent Domains: A Weighted Constrained Potential Dynamic Game Approach
· IEEE Transactions on Robotics
Robotics · 7 min text · 55 citations
-
Control-Ready Uncertainty for Trajectory Diffusion
Preprint posted · CoRL 2026 (accepted)
Robotics · 22 min text · 47 citations
-
RoboKit: Accelerating Parallel Differentiable Optimization for Robot Learning
Conference session · CoRL 2026 (accepted)
Robotics · 12 min text · 81 citations
-
FingerEye: Learning Dexterous Manipulation with Continuous Vision-Tactile Sensing
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 56 citations
-
EagleVLA: Towards Onboard Real-Time Robot Control via Foresight-Aligned Asynchronous Inference
Preprint posted · CoRL 2026 (accepted)
Robotics · 7 min text · 60 citations
-
ResSafe: Learning Safety Filtering with Residual Reinforcement Learning for Humanoids
Preprint posted · CoRL 2026 (accepted)
Robotics · 7 min text · 55 citations
-
Driving, Fast or Slow? Neuro-Symbolic Guidance for Motion Prediction in Multi-Modal Ground Mobility
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 50 citations
-
TAM: Torque Adaptation Module for Robust Motion Transfer in Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 19 min text · 46 citations
-
MolmoB0T: Large-Scale Simulation Enables Zero-Shot Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 16 min text · 54 citations
-
TRIAD Planner: Three-Level Robust Imitation Learning for Autonomous Driving
Conference session · CoRL 2026 (accepted)
Robotics · 9 min text · 27 citations
-
Bayesian Active Learning for Intent Disambiguation in Interactive Robot Planning
Preprint posted · CoRL 2026 (accepted)
Robotics · 16 min text · 28 citations
-
MosaicIMU: Composing Carrier Experts for Generalizable Neural Inertial Odometry
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 35 citations
-
From Region Arrival to Instance-Level Grounding in Vision-and-Language Navigation
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 34 citations
-
WireCraft: A Simulation Benchmark for Industrial DLO Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 58 citations
-
When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 66 citations
-
VisionSIM: Beyond Idealized Sensors — A Multi-Modal Simulator for Robot Perception
Preprint posted · CoRL 2026 (accepted)
Robotics · 14 min text · 101 citations
-
SidewalkBench: Benchmarking Visual Navigation on Urban Sidewalks
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 62 citations
-
PHUMA: Physically Reliable Humanoid Locomotion Dataset
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 86 citations
-
Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 55 citations
-
ATOM-Bench: A Real-World Benchmark for Atomic Skills and Compositional Generalization in Manipulation Policies
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 37 citations
-
What Are We Actually Benchmarking in Robot Manipulation?
Preprint posted · CoRL 2026 (accepted)
Robotics · 16 min text · 63 citations
-
GRAIL: Generating Humanoid Loco-Manipulation from World Models
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 86 citations
-
Scalable Deep Unfolding of Conic Optimizers
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 50 citations
-
Playful Agentic Robot Learning
Preprint posted · CoRL 2026 (accepted)
Robotics · 16 min text · 61 citations
-
OSCAR: Obstacle Survival Curves for Adaptive Robot Navigation
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 31 citations
-
Meta-Ctrl: Guaranteed Plan Generation by Decoupling Syntactic and Semantic Constraints
Preprint posted · CoRL 2026 (accepted)
Robotics · 17 min text · 26 citations
-
Leave It Better Than Found: Value-Guided Cooperation in Sequential Tasks
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 37 citations
-
Diffusion ReRoll: Revisable Denoising for Robotic Sequential Prediction
Preprint posted · CoRL 2026 (accepted)
Robotics · 22 min text · 41 citations
-
Collaborative Planning for UAV-UGV Teams under Operational Constraints via Masked Error-bounded Neuro-Symbolic Diffusion
Preprint posted · CoRL 2026 (accepted)
Robotics · 12 min text · 75 citations
-
Long-Horizon Multi-Object Rearrangement via Distilled Tree Search over Continuous Actions
Preprint posted · CoRL 2026 (accepted)
Robotics · 14 min text · 97 citations
-
Visible Touch: Rendering Contact for Visuomotor Policies
Preprint posted · CoRL 2026 (accepted)
Robotics · 17 min text · 37 citations
-
SERF: Spatiotemporal Environment and Robot Feature Map for Long-Horizon Mobile Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 8 min text · 40 citations
-
Semantic Anchoring for Robotic Action Representations
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 46 citations
-
RAM: Reachability Across Morphologies
Preprint posted · CoRL 2026 (accepted)
Robotics · 12 min text · 63 citations
-
Noise-Invariant Spatial Encoding for 3D Manipulation Policies under Depth Corruption
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 42 citations
-
Never Too Late for Force: Accelerating VLA Post-Training with Reactive Force Injection
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 32 citations
-
DeTac: Dual-branch Decoupling of Texture and Force Signals in Vision-Based Tactile Sensors
Preprint posted · CoRL 2026 (accepted)
Robotics · 7 min text · 65 citations
-
DAVATaR: Depth-Aware Visually Augmented Tactile Learning Across the Air-Water Interface
Conference session · CoRL 2026 (accepted)
Robotics · 9 min text · 35 citations
-
Articulated 3D Scene Graphs for Open-World Mobile Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 60 citations
-
Anchor3R: Streaming 3D Reconstruction with Transient Anchors for Long-Horizon Visual Mapping
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 57 citations
-
TEMPO: Learning Temporal Context for Dynamic Robot Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 38 citations
-
Targeting World Models to Compromise Robot Learning Pipelines
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 61 citations
-
World Action Models are Zero-shot Policies
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 92 citations
-
Mem-World: Memory-Augmented Action-Conditioned World Models for Persistent Robot Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 5 min text · 39 citations
-
ZETA: A Controlled Study of Zero-Shot Cross-Embodiment VLA Transfer for Tabletop Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 73 citations
-
PRTS: A Primitive Reasoning and Tasking System via Contrastive Representations
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 73 citations
-
MoRE: Mixture-of-Experts for Multi-Physics Robotic World Models
Conference session · CoRL 2026 (accepted)
Robotics · 9 min text · 74 citations
-
MolmoAct2: Action Reasoning Models for Real-World Deployment
Preprint posted · CoRL 2026 (accepted)
Robotics · 23 min text · 94 citations
-
MEM: Multi-Scale Embodied Memory for Vision Language Action Models
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 55 citations
-
Learning What to Say to Your VLA: Mostly Harmless Vision Language Action Model Steering
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 41 citations
-
π₀.₇: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 112 citations
-
RoboTTT: Context Scaling for Robot Policies
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 83 citations
-
EgoScale: Scaling Dexterous Manipulation with Diverse Egocentric Human Data
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 46 citations
-
Being-H0.5: Scaling Human-Centric Robot Learning for Cross-Embodiment Generalization
Preprint posted · CoRL 2026 (accepted)
Robotics · 17 min text · 140 citations
-
OpenHLM: An Empirical Recipe for Whole-Body Humanoid Loco-Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 54 citations
-
LadderMan: Learning Humanoid Perceptive Ladder Climbing
Preprint posted · CoRL 2026 (accepted)
Robotics · 8 min text · 42 citations
-
Humanoid Navigation in Cluttered Environments with a Whole-Body Vision-Language-Action Model
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 55 citations
-
CMP: Robust Whole-Body Tracking for Loco-Manipulation via Competence Manifold Projection
Preprint posted · CoRL 2026 (accepted)
Robotics · 14 min text · 58 citations
-
SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 29 citations
-
RoboNaldo: Accurate, Stable and Powerful Humanoid Soccer Shooting via Motion-Guided Curriculum Reinforcement Learning
Preprint posted · CoRL 2026 (accepted)
Robotics · 12 min text · 32 citations
-
TARC: Time-Adaptive Robotic Control
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 35 citations
-
Shape Your Body: Value Gradients for Multi-Embodiment Robot Design
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 62 citations
-
SCoCaT: Success Conditioned Constrained Reinforcement Learning for Spacecraft Docking
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 30 citations
-
Learning Loco-Manipulation From SMPC Demonstrations With Sparse Offline-to-Online RL
Preprint posted · CoRL 2026 (accepted)
Robotics · 8 min text · 27 citations
-
Kamino: GPU-based Simulation of Rigid Multi-Body Physics with Kinematic Loops
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 39 citations
-
FedGuide: Diffusion Prior Alignment and Value Baseline Guidance for Heterogeneous Federated Reinforcement Learning
Preprint posted · CoRL 2026 (accepted)
Robotics · 17 min text · 52 citations
-
Emergent Control via Prior-Based Free Energy Minimization
Conference session · CoRL 2026 (accepted)
Robotics · Reading note
-
Chunk-wise Model-based Policy Optimization
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 42 citations
-
RL Token: Bootstrapping Online RL with Vision-Language-Action Models
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 42 citations
-
Play2Perfect: What Matters in Dexterous Play Pretraining for Precise Assembly?
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 66 citations
-
Flow Policy Gradients for Robot Control
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 68 citations
-
Robots that Collaborate: Sequential Asymmetric Imitation for Learning Coupled Robot Policies
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 34 citations
-
FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 30 citations
-
Efficient and Reliable Teleoperation through Real-to-Sim-to-Real Shared Autonomy
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 81 citations
-
Diagnosing Compositional Generalization in Sequential Robot Tasks
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 28 citations
-
PACE: Temporal-Role Tactile Conditioning for Chunk-Based Insertion Policies
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 54 citations
-
FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning
Preprint posted · CoRL 2026 (accepted)
Robotics · 12 min text · 51 citations
-
Diversity You Can Actually Measure: A Fast, Model-Free Diversity Metric for Robotics Datasets
Preprint posted · CoRL 2026 (accepted)
Robotics · 13 min text · 80 citations
-
WM-Craftnet: World Synesthesia Model for Robust and Generalizable Dexterous In-Hand Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 25 citations
-
Robustness without Wrinkles: Parallel Simulation and Robust MPC for Certified Deformable Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 61 citations
-
REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models
Preprint posted · CoRL 2026 (accepted)
Robotics · 12 min text · 71 citations
-
Learning Physics-Guided Residual Dynamics for Deformable Object Simulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 8 min text · 98 citations
-
Learning Human Intention for Robotic Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 86 citations
-
KPGrasp: Scalable Keypoint Flow Matching for Dexterous Grasp Generation
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 54 citations
-
GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 74 citations
-
Enabling Robust Cloth Manipulation via Inference-Time Simulator-in-the-Loop Refinement
Preprint posted · CoRL 2026 (accepted)
Robotics · 14 min text · 69 citations
-
EmbodiSteer: Steering Embodiment-Agnostic Visuomotor Policies with Joint-Space Guidance for Zero-Shot Cross-Embodiment Deployment
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 59 citations
-
Demystifying When and Why VLAs Fail in Contact-Rich Tasks and How to Fix Them
Preprint posted · CoRL 2026 (accepted)
Robotics · 14 min text · 27 citations
-
Constraint-Decomposed Impedance Control from Generative Policy for Articulated Object Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 20 min text · 22 citations
-
Combining Constrained Sampling and Reinforcement Learning for Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 72 citations
-
BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 8 min text · 56 citations
-
BiDexGrasp: Coordinated Bimanual Dexterous Grasps across Object Geometries and Sizes
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 49 citations
-
OPENDEXGRASP: Open-vocabulary Task-Oriented Dexterous Grasping
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 41 citations
-
Why Automate This? Exploring the Connection between Time Use, Well-being and Robot Automation Across Social Groups
· IASEAI 2026
Robotics · 8 min text · 60 citations
-
BiFlex: A Passive Bimodal Stiffness Flexible Wrist for Manipulation in Unstructured Environments
· IEEE Robotics and Automation Letters
Robotics · 5 min text · 26 citations
-
DataMIL: Selecting Data for Robot Imitation Learning with Datamodels
· ICLR 2026
Robotics · 10 min text · 57 citations
-
SLAC: Safe and Efficient Real-Robot Reinforcement Learning via Unsupervised Simulation Pre-Training
· CoRL 2025
Robotics · 11 min text · 76 citations
-
CASPER: Inferring Diverse Intents for Assistive Teleoperation with Vision Language Models
· CoRL 2025
Robotics · 7 min text · 74 citations
-
SafeMimic: Towards Safe and Autonomous Human-to-Robot Imitation for Mobile Manipulation
· Robotics: Science and Systems 2025
Robotics · 9 min text · 68 citations
-
RoboArena: Distributed Real-World Evaluation of Generalist Robot Policies
· CoRL 2025
Robotics · 12 min text · 73 citations
-
COLLAGE: Adaptive Fusion-based Retrieval for Augmented Policy Learning
· CoRL 2025
Robotics · 8 min text · 54 citations
-
Mixed-Initiative Dialog for Human-Robot Collaborative Manipulation
· ICRA 2026
Robotics · 11 min text · 93 citations
-
MimicDroid: In-Context Learning for Humanoid Robot Manipulation from Human Play Videos
· ICRA 2026
Robotics · 10 min text · 59 citations
-
RoboSSM: Scalable In-context Imitation Learning via State-Space Models
· IROS 2026
Robotics · 6 min text · 37 citations
-
Mash, Spread, Slice! Learning to Manipulate Object States via Visual Spatial Progress
· ICRA 2026
Robotics · 5 min text · 53 citations
-
MoMaGen: Generating Demonstrations under Soft and Hard Constraints for Multi-Step Bimanual Mobile Manipulation
· ICLR 2026
Robotics · 11 min text · 54 citations
-
GeT-USE: Learning Generalized Tool Usage for Bimanual Mobile Manipulation via Simulated Embodiment Extensions
· IROS 2025
Robotics · 7 min text · 29 citations
-
CAVER: Curious Audiovisual Exploring Robot
· ICRA 2026
Robotics · 6 min text · 51 citations
-
Searching in Space and Time: Unified Memory-Action Loops for Open-World Object Retrieval
· ICRA 2026
Robotics · 9 min text · 55 citations
-
Factored Latent Action World Models
· ICML 2026
World models · 10 min text · 38 citations
-
Large-Language-Model-Guided State Estimation for Partially Observable Task and Motion Planning
· ICRA 2026
Robotics · 10 min text · 35 citations
-
Simple Recipe Works: Vision-Language-Action Models are Natural Continual Learners with Reinforcement Learning
· RLC 2026
Robotics · 13 min text · 66 citations
-
Rapid Adaptation of Particle Dynamics for Generalized Deformable Object Mobile Manipulation
· ICRA 2026
Robotics · 8 min text · 52 citations
-
VOFA: Visual Object Goal Pushing with Force-Adaptive Control for Humanoids
· IROS 2026
Robotics · 9 min text · 27 citations
-
EgoExo-WM: Unlocking Exo Video for Ego World Models
Preprint posted · arXiv (preprint)
World models · 11 min text · 81 citations
-
Understanding the Impact of Geometric Foundation Models on Vision-Language-Action Models
Preprint posted · arXiv (preprint)
World models · 12 min text · 55 citations
-
Scaling Short-Term Memory of Visuomotor Policies for Long-Horizon Tasks
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 63 citations
-
Memory Retrieval in Visuomotor Policies for Long-Horizon Robot Control
· Robotics: Science and Systems 2026
Robotics · 8 min text · 48 citations
-
CoDex: Learning Compositional Dexterous Functional Manipulation without Demonstrations
· ICRA 2026
Robotics · 5 min text · 48 citations
-
OopsieVerse: A Safety Benchmark with Damage-Aware Simulation for Robot Manipulation
· Robotics: Science and Systems 2026
Robotics · 11 min text · 38 citations
-
Beyond Episodic Evaluation: Memory Architectural Bottlenecks in Sequential Embodied Question Answering
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 27 citations
-
SimEX: Simulation-Integrated Robotics AutoResearch
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 50 citations
-
ReDex: Repairing Sim-to-Real Dexterous Policies by Finger-Level Compliant Interaction
Preprint posted · arXiv (preprint)
Robotics · 5 min text · 33 citations
-
GUILD: Guided Iterative Data Collection for Imitation Learning
Preprint posted · CoRL 2026 (accepted)
Robotics · 12 min text · 59 citations
-
MOMASPLAT: Dynamic Novel-View Demonstration Augmentation for Mobile Manipulation
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 57 citations
-
OopsieRecover: Learning Transferable VLA Corrections from Semantic Interventions
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 55 citations
-
DDAT: Diffusion Policies Enforcing Dynamically Admissible Robot Trajectories
· Robotics: Science and Systems 2025
Robotics · 12 min text · 55 citations
-
Leveraging Large Language Models for Effective and Explainable Multi-Agent Credit Assignment
· AAMAS 2025
Robotics · 7 min text · 62 citations
-
RAMEN: Real-time Asynchronous Multi-agent Neural Implicit Mapping
· Robotics: Science and Systems 2025
Robotics · 6 min text · 52 citations
-
UDON: Uncertainty-weighted Distributed Optimization for Multi-Robot Neural Implicit Mapping under Extreme Communication Constraints
· ICRA 2026
Robotics · 4 min text · 26 citations
-
MIMIC-D: Multi-modal Imitation for MultI-agent Coordination with Decentralized Diffusion Policies
· ICRA 2026
Robotics · 10 min text · 38 citations
-
CRAFT: Coaching Reinforcement Learning Autonomously using Foundation Models for Multi-Robot Coordination Tasks
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 33 citations
-
TACO: Temporal Consensus Optimization for Continual Neural Mapping
· Robotics: Science and Systems 2026
Robotics · 8 min text · 59 citations
-
Matching Multiple Experts: On the Exploitability of Multi-Agent Imitation Learning
· ICLR 2026
Robotics · 13 min text · 50 citations
-
Learning Control Policies to Provably Satisfy Hard Affine Constraints for Black-Box Hybrid Dynamical Systems
Preprint posted · IEEE Conference on Decision and Control 2026 (accepted)
Robotics · 12 min text · 41 citations
-
X-Safe: Embodiment-Agnostic Safeguarding for Robot Arm Manipulation Policies via Action Masking
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 39 citations
-
Scaling Nonlinear Optimization: Many Problems One GPU
Preprint posted · arXiv (preprint)
Robotics · 5 min text · 38 citations
-
Congestion Games with Heterogeneous Valuations: An Optimal Transport Approach
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 51 citations
-
HuGo: LLMs as Whole-Body Policy Code Designers for Humanoid Loco-Manipulation
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 32 citations
-
SAMBAR: Selective Anchoring via Method of Multipliers for Balanced Knowledge Acquisition and Retention in Vision-Language-Action Models
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 36 citations
-
Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand
Preprint posted · arXiv (preprint)
Robotics · 8 min text · 30 citations
-
PopGen: Generating Semantically Diverse Populations for Zero-Shot Coordination
Preprint posted · arXiv (preprint)
Robotics · 16 min text · 46 citations
-
CusADi2: Latency-Optimized Batch Computation of Robot Kinematics, Jacobians, and Hessians on the GPU
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 27 citations
-
A Robust Approach to Belief Mismatch in Non-cooperative Dynamic Games
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 29 citations
-
A Unit-based System and Dataset for Expressive Direct Speech-to-Speech Translation
Preprint posted · arXiv (preprint)
Speech Recognition · 6 min text · 32 citations
-
When End-to-End is Overkill: Rethinking Cascaded Speech-to-Text Translation
Preprint posted · arXiv (preprint)
Speech Recognition · 5 min text · 32 citations
-
Embrace Contacts: humanoid shadowing with full body ground contacts
· CoRL 2025
Robotics · 9 min text · 55 citations
-
Intrinsic Entropy of Context Length Scaling in LLMs
· ICLR 2026
Language models · 15 min text · 36 citations
-
VR-Robo: A Real-to-Sim-to-Real Framework for Visual Robot Navigation and Locomotion
· IEEE Robotics and Automation Letters
Robotics · 6 min text · 42 citations
-
Chameleon: Fast-Slow Neuro-Symbolic Lane Topology Extraction
· ICRA 2025
Robotics · 7 min text · 37 citations
-
TrackOcc: Camera-Based 4D Panoptic Occupancy Tracking
· ICRA 2025
Computer Vision · 9 min text · 42 citations
-
MoE-Loco: Mixture of Experts for Multitask Locomotion
· IROS 2025
Robotics · 7 min text · 54 citations
-
Learning Personalized Driving Styles via Reinforcement Learning from Human Feedback
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 42 citations
-
RoboEngine: Plug-and-Play Robot Data Augmentation with Semantic Robot Segmentation and Background Generation
· IROS 2025
Robotics · 9 min text · 42 citations
-
Conditioning Matters: Training Diffusion Policies is Faster Than You Think
· NeurIPS 2025
Robotics · 11 min text · 38 citations
-
Challenger: Affordable Adversarial Driving Video Generation for Safety Testing
· IASEAI 2026
Robotics · 9 min text · 70 citations
-
Diffusion-Based Generative Models for 3D Occupancy Prediction in Autonomous Driving
· ICRA 2025
Computer Vision · 8 min text · 44 citations
-
Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models
· NeurIPS 2025
Robotics · 8 min text · 88 citations
-
ORV: 4D Occupancy-centric Robot Video Generation
· CVPR 2026
Robotics · 12 min text · 137 citations
-
BEV-VAE: Multi-view Image Generation with Spatial Consistency for Autonomous Driving
Preprint posted · arXiv (preprint)
Computer Vision · 10 min text · 25 citations
-
Re:Form - Reducing Human Priors in Scalable Formal Software Verification with RL in LLMs: A Preliminary Study on Dafny
· Transactions on Machine Learning Research
Reasoning · 17 min text · 95 citations
-
Morpheus: A Neural-driven Animatronic Face with Hybrid Actuation and Diverse Emotion Control
· Robotics: Science and Systems 2025
Robotics · 9 min text · 57 citations
-
Reusing Attention for One-stage Lane Topology Understanding
· IROS 2025
Robotics · 6 min text · 55 citations
-
LONG3R: Long Sequence Streaming 3D Reconstruction
· ICCV 2025
Computer Vision · 9 min text · 89 citations
-
CRUISE: Cooperative Reconstruction and Editing in V2X Scenarios using Gaussian Splatting
· IROS 2025
Robotics · 6 min text · 83 citations
-
Delving into Mapping Uncertainty for Mapless Trajectory Prediction
· IROS 2025
Robotics · 7 min text · 42 citations
-
GS-Occ3D: Scaling Vision-Only Occupancy Reconstruction with Gaussian Splatting
· ICCV 2025
Computer Vision · 9 min text · 109 citations
-
DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking
· ICLR 2026
Robotics · 14 min text · 54 citations
-
Galaxea Open-World Dataset and G0 Dual-System VLA Model
· ICRA 2026
Robotics · 10 min text · 30 citations
-
OccVLA: Vision-Language-Action Model with Implicit 3D Occupancy Supervision
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 46 citations
-
SLAM-Former: Putting SLAM Into One Transformer
· ECCV 2026
Computer Vision · 12 min text · 47 citations
-
Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving
· ICLR 2026
Robotics · 10 min text · 40 citations
-
DepthVLA: Enhancing Vision-Language-Action Models with Depth-Aware Spatial Reasoning
· ICRA 2026
Robotics · 5 min text · 48 citations
-
DAP: A Discrete-token Autoregressive Planner for Autonomous Driving
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 49 citations
-
RoboCOIN: An Open-Sourced Bimanual Robotic Data Collection for Integrated Manipulation
Preprint posted · arXiv (preprint)
Robotics · 10 min text · 60 citations
-
FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via Neural Action Tokenization
· ICLR 2026
Robotics · 14 min text · 68 citations
-
Deep Whole-body Parkour
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 28 citations
-
Hiking in the Wild: A Scalable Perceptive Legged Parkour Framework for Humanoid Robots
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 54 citations
-
TTT-Parkour: Rapid Test-Time Training for Perceptive Legged Humanoid Parkour
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 60 citations
-
ActionCodec: What Makes for Good Action Tokenizers
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 41 citations
-
Fast-WAM: Do World Action Models Need Test-time Future Imagination?
Preprint posted · NeurIPS 2026 (accepted)
Robotics · 8 min text · 38 citations
-
VolumeDP: Modeling Volumetric Representation for Manipulation Policy Learning
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 42 citations
-
Generative Control as Optimization: Time Unconditional Flow Matching for Adaptive and Robust Robotic Control
Preprint posted · NeurIPS 2026 (accepted)
Robotics · 14 min text · 48 citations
-
UniDex: A Robot Foundation Suite for Universal Dexterous Hand Control from Egocentric Human Videos
· CVPR 2026
Robotics · 11 min text · 78 citations
-
Complet4R: Geometric Complete 4D Reconstruction
· CVPR 2026
Computer Vision · 11 min text · 58 citations
-
Dexora: Open-source VLA for High-DoF Bimanual Dexterity
· ICRA 2026
Robotics · 8 min text · 44 citations
-
DriveMA: Driving Vision-Language-Action Models with verifiable Meta-Actions
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 31 citations
-
Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models
Preprint posted · CoRL 2026 (accepted)
Robotics · 10 min text · 41 citations
-
OMG: Omni-Modal Motion Generation for Generalist Humanoid Control
Preprint posted · arXiv (preprint)
Robotics · 16 min text · 74 citations
-
TopoRetarget: Interaction-Preserving Retargeting for Dexterous Manipulation
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 31 citations
-
Is Your Trajectory Displacement Safe in Long-tail?
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 34 citations
-
Orca: The World is in Your Mind
Preprint posted · arXiv (preprint)
Robotics · 20 min text · 95 citations
-
Agent-driven Long-tail Simulation for Autonomous Driving
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 55 citations
-
Automated Synthesis of Facial Mechanisms for Conversational Animatronic Robots
· Robotics: Science and Systems 2026
Robotics · 18 min text · 118 citations
-
SLAMFormer-∞: Infinite SLAM Transformer for Unbounded Frontend and Backend Processing
Preprint posted · arXiv (preprint)
Computer Vision · 6 min text · 41 citations
-
Sparse Weight Decomposition for Efficient Circuit Extraction
Preprint posted · arXiv (preprint)
Interpretability · 18 min text · 47 citations
-
G0.5: One Autoregressive Stream for Robot Reasoning and Action
Preprint posted · arXiv (preprint)
Robotics · 14 min text · 61 citations
-
OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining
Preprint posted · arXiv (preprint)
World models · 17 min text · 108 citations
-
SkelWAM: A Skeleton-Guided World-Action Model for Zero-Shot Cross-Embodiment Manipulation
Preprint posted · arXiv (preprint)
Robotics · 6 min text · 35 citations
-
MSK-Bench: Benchmarking Full-Body Musculoskeletal Motor Control Across Tasks, Control Paradigms, and Physiological Metrics
Preprint posted · arXiv (preprint)
Robotics · 12 min text · 127 citations
-
TactileStep: Sole Tactile Learning for Regulating Foot–Terrain Interaction in Humanoid Locomotion
Preprint posted · CoRL 2026 (accepted)
Robotics · 11 min text · 47 citations
-
Echo in the Steps: Learning Perceptive Humanoid Parkour with Gated Memory
Preprint posted · CoRL 2026 (accepted)
Robotics · 9 min text · 46 citations
-
Representation World Model: Learning States, Transition and Executable Plans in Representation
Preprint posted · arXiv (preprint)
World models · 9 min text · 35 citations
-
World SLAM Model: Joint World Modeling for SLAM and Navigation
Preprint posted · arXiv (preprint)
Computer Vision · 9 min text · 37 citations
-
WB-WAM: Heterogeneous Body-Hand Pre-training for Humanoid Loco-Manipulation
Preprint posted · arXiv (preprint)
Robotics · 9 min text · 43 citations
-
Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching
Preprint posted · NeurIPS 2026 (accepted)
Robotics · 10 min text · 40 citations
-
Supervising Sound Localization by In-the-wild Egomotion
· CVPR 2025 Spotlight
Audio and acoustics · 10 min text · 63 citations
-
DexNinja: Learning Robust Dexterous Cutting Policy with a Real-to-Sim-to-Real Data Engine
Preprint posted · CoRL 2026 (accepted)
Robotics · 14 min text · 88 citations
-
EqC: Equilibrium Energy Control for Introspective Lifelong Robot Learning
Preprint posted · arXiv (preprint)
Robotics · 20 min text · 37 citations
-
Proving and Improving the Platonic Representation Hypothesis with JEPAs
Preprint posted · arXiv (preprint)
Learning theory · 20 min text · 75 citations
-
Robotic Data Inoculation
Preprint posted · arXiv (preprint)
Robotics · 7 min text · 31 citations
-
Agent as Policy for Robotic Manipulation
Preprint posted · arXiv (preprint)
Robotics · 13 min text · 32 citations
-
CS231n: Convolutional Neural Networks for Visual Recognition
Course notes introduced · Stanford University (maintained course notes)
Deep Learning Foundations · 4 min text · 150 citations
-
Kolmogorov Complexity and Algorithmic Randomness — Algorithmic Statistics and Appendices
Published · AMS · Mathematical Surveys and Monographs 220
Information Theory · 13 min text · 233 citations
-
Machine Super Intelligence
Dissertation dated · University of Lugano (doctoral dissertation)
Information Theory and Intelligence · 6 min text · 188 citations
-
A Tutorial Introduction to the Minimum Description Length Principle
· Advances in Minimum Description Length: Theory and Applications, MIT Press (chapters 1–2)
Learning theory · 16 min text · 93 citations
-
Scaling Laws for Neural Language Models
Preprint posted · arXiv (preprint)
Language models · 10 min text · 50 citations
-
Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
· ICML 2016
Speech Recognition · 7 min text · 70 citations
-
Neural Turing Machines
Preprint posted · arXiv (preprint)
Sequence Modeling · 5 min text · 42 citations
-
Quantifying the Rise and Fall of Complexity in Closed Systems: The Coffee Automaton
Preprint posted · arXiv (preprint)
Learning theory · 5 min text · 13 citations
-
Relational recurrent neural networks
Published · NeurIPS 2018
Reasoning · 5 min text · 47 citations
-
Variational Lossy Autoencoder
· ICLR 2017
Learning theory · 6 min text · 62 citations
-
A simple neural network module for relational reasoning
Published · NeurIPS 2017
Reasoning · 6 min text · 48 citations
-
Identity Mappings in Deep Residual Networks
Published · ECCV 2016
Computer Vision · 4 min text · 23 citations
-
Neural Machine Translation by Jointly Learning to Align and Translate
· ICLR 2015
Sequence Modeling · 5 min text · 29 citations
-
Attention Is All You Need
· NeurIPS 2017
Sequence Modeling · 5 min text · 40 citations
-
Neural Message Passing for Quantum Chemistry
Published · ICML 2017
Graph Learning · 5 min text · 37 citations
-
Multi-Scale Context Aggregation by Dilated Convolutions
Published · ICLR 2016
Computer Vision · 4 min text · 35 citations
-
Deep Residual Learning for Image Recognition
Published · CVPR 2016
Computer Vision · 4 min text · 50 citations
-
GPipe: Easy Scaling with Micro-Batch Pipeline Parallelism
Published · NeurIPS 2019
Efficient Learning · 4 min text · 65 citations
-
Order Matters: Sequence to Sequence for Sets
· ICLR 2016
Sequence Modeling · 4 min text · 23 citations
-
ImageNet Classification with Deep Convolutional Neural Networks
Published · NeurIPS 2012
Computer Vision · 4 min text · 26 citations
-
Pointer Networks
· NeurIPS 2015
Sequence Modeling · 4 min text · 20 citations
-
Keeping Neural Networks Simple by Minimizing the Description Length of the Weights
· COLT 1993
Learning theory · 4 min text · 7 citations
-
Recurrent Neural Network Regularization
Preprint posted · arXiv (preprint)
Sequence Modeling · 3 min text · 34 citations
-
Understanding LSTM Networks
Published · Christopher Olah’s blog
Deep Learning Foundations · 2 min text · 23 citations
-
The Unreasonable Effectiveness of Recurrent Neural Networks
Published · Andrej Karpathy’s blog
Deep Learning Foundations · 3 min text · 68 citations
-
The First Law of Complexodynamics
Published · Shtetl-Optimized (research essay)
Deep Learning Foundations · 2 min text · 11 citations
-
The Annotated Transformer
Published · Harvard NLP (annotated notebook)
Deep Learning Foundations · 5 min text · 18 citations
-
UltraBERT: Beyond the Limit of BERT Pre-Training
Preprint posted · arXiv (preprint)
Language models · 12 min text · 101 citations
-
LeWAM: End-to-End World and Action Modeling with JEPAs
Preprint posted · arXiv (preprint)
World models · 11 min text · 53 citations
-
GreenWAM: A World Foundation Model for Robotics
Preprint posted · arXiv (preprint)
World models · 16 min text · 106 citations
-
Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards
Preprint posted · arXiv (preprint)
AI agents · 14 min text · 85 citations
-
VideoJev: System One Model for Video Understanding and Interaction
Preprint posted · arXiv (preprint)
Multimodal reasoning · 10 min text · 63 citations
-
Jeva: System One Models from First Principles
Preprint posted · arXiv (preprint)
Language models · 12 min text · 27 citations
-
The Geometry of Empowerment
Preprint posted · arXiv (preprint)
Machine learning · 15 min text · 48 citations
-
Behavioral Cloning Mystery
Preprint posted · arXiv (preprint)
Robotics · 11 min text · 56 citations
-
VISTA: A Visual Harness for Reasoning in an Interactive World
Preprint posted · arXiv (preprint)
AI agents · 10 min text · 64 citations
-
Training Object Permanence in World Models
Preprint posted · arXiv (preprint)
World models · 10 min text · 45 citations
-
Natural Image Pretraining Improves Abstract Reasoning
· ECCV 2026
Reasoning · 7 min text · 51 citations
-
Evaluating Sparse Autoencoders on Targeted Concept Erasure Tasks
· arXiv (preprint)
Interpretability · 6 min text · 31 citations
-
Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning
· Mechanistic Interpretability Workshop at NeurIPS 2025
Interpretability · 12 min text · 92 citations
-
Open Problems in Mechanistic Interpretability
Published · Transactions on Machine Learning Research
Interpretability · 7 min text · 442 citations
-
Refusal in Language Models Is Mediated by a Single Direction
· NeurIPS 2024
Interpretability · 12 min text · 99 citations
-
Characterizing interference weights in a tiny language model
Published · Transformer Circuits Thread
Interpretability · 8 min text · 56 citations
-
Would this change your answer? Evaluating Explanations of LLM Behavior In The Wild with Counterfactual Experiments
· arXiv (preprint)
Interpretability · 10 min text · 35 citations
-
Verbalizable Representations Form a Global Workspace in Language Models
Published · Transformer Circuits Thread
Interpretability · 19 min text · 171 citations
-
Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
Published · Transformer Circuits Thread
Interpretability · 8 min text · 59 citations
-
Introspection Adapters: Training LLMs to Report Their Learned Behaviors
Preprint posted · ICML 2026 (accepted)
Interpretability · 9 min text · 34 citations
-
Emotion Concepts and their Function in a Large Language Model
Published · Transformer Circuits Thread
Interpretability · 16 min text · 53 citations
-
AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors
· arXiv (preprint)
Interpretability · 9 min text · 39 citations
-
Three Concrete Challenges and Two Hopes for the Safety of Unsupervised Elicitation
· arXiv (preprint)
Interpretability · 8 min text · 51 citations
-
Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs
· arXiv (preprint)
Interpretability · 20 min text · 31 citations
-
The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
· arXiv (preprint)
Interpretability · 12 min text · 37 citations
-
Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers
· arXiv (preprint)
Interpretability · 11 min text · 33 citations
-
Evaluating honesty and lie detection techniques on a diverse suite of dishonest models
· Alignment Science Blog
Interpretability · 8 min text · 30 citations
-
Emergent Introspective Awareness in Large Language Models
Published · Transformer Circuits Thread
Interpretability · 8 min text · 43 citations
-
When Models Manipulate Manifolds: The Geometry of a Counting Task
Published · Transformer Circuits Thread
Interpretability · 7 min text · 79 citations
-
Believe It or Not: How Deeply do LLMs Believe Implanted Facts?
· arXiv (preprint)
Interpretability · 11 min text · 54 citations
-
A Toy Model of Mechanistic (Un)Faithfulness
Published · Transformer Circuits Thread
Interpretability · 3 min text · 4 citations
-
Tracing Attention Computation Through Feature Interactions
Published · Transformer Circuits Thread
Interpretability · 4 min text · 8 citations
-
persona vectors: monitoring and controlling character traits in language models
· arXiv (preprint)
Interpretability · 13 min text · 62 citations
-
A Toy Model of Interference Weights
Published · Transformer Circuits Thread
Interpretability · 4 min text · 7 citations
-
Sparse mixtures of linear transforms
Published · Transformer Circuits Thread
Interpretability · 3 min text · 8 citations
-
Building and evaluating alignment auditing agents
Published · Anthropic Alignment Science Blog
Interpretability · 7 min text · 23 citations
-
Progress on Attention
Published · Transformer Circuits Thread
Interpretability · 3 min text · 7 citations
-
On the Biology of a Large Language Model
Published · Transformer Circuits Thread
Interpretability · 11 min text · 95 citations
-
Circuit Tracing: Revealing Computational Graphs in Language Models
Published · Transformer Circuits Thread
Interpretability · 11 min text · 144 citations
-
Auditing Language Models for Hidden Objectives
· arXiv (preprint)
Interpretability · 19 min text · 60 citations
-
SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
· ICML 2025
Interpretability · 12 min text · 33 citations
-
Insights on Crosscoder Model Diffing
Published · Transformer Circuits Thread
Interpretability · 2 min text · 8 citations
-
Stage-Wise Model Diffing
· Transformer Circuits Thread
Interpretability · 3 min text · 2 citations
-
Sparse Crosscoders for Cross-Layer Features and Model Diffing
· Transformer Circuits Thread
Interpretability · 5 min text · 43 citations
-
Evaluating feature steering: A case study in mitigating social biases
· Anthropic research report
Interpretability · 5 min text · 11 citations
-
Using Dictionary Learning Features as Classifiers
· Transformer Circuits Thread
Interpretability · 3 min text · 10 citations
-
Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet
· Transformer Circuits Thread
Interpretability · 11 min text · 77 citations
-
Simple probes can catch sleeper agents
· Anthropic Alignment Note (preliminary)
Interpretability · 4 min text · 16 citations
-
Steering Llama 2 via Contrastive Activation Addition
· ACL 2024
Interpretability · 7 min text · 28 citations
-
Towards Monosemanticity: Decomposing Language Models With Dictionary Learning
· Transformer Circuits Thread
Interpretability · 15 min text · 75 citations
-
Studying Large Language Model Generalization with Influence Functions
· arXiv (preprint)
Interpretability · 12 min text · 111 citations
-
Privileged Bases in the Transformer Residual Stream
· Transformer Circuits Thread
Interpretability · 3 min text · 4 citations
-
Superposition, Memorization, and Double Descent
· Transformer Circuits Thread
Interpretability · 5 min text · 13 citations
-
Engineering Monosemanticity in Toy Models
· arXiv (preprint)
Interpretability · 5 min text · 7 citations
-
Toy Models of Superposition
· Transformer Circuits Thread
Interpretability · 14 min text · 71 citations
-
Softmax Linear Units
· Transformer Circuits Thread
Interpretability · 7 min text · 70 citations
-
Scaling Laws and Interpretability of Learning from Repeated Data
· arXiv (preprint)
Interpretability · 6 min text · 34 citations
-
In-context Learning and Induction Heads
· Transformer Circuits Thread
Interpretability · 12 min text · 51 citations
-
A Mathematical Framework for Transformer Circuits
· Transformer Circuits Thread
Interpretability · 12 min text · 79 citations
-
Passive Interaction Control With Dynamical Systems
· IEEE Robotics and Automation Letters
Robotics · 7 min text · 21 citations
-
Model predictive optimization for imitation learning from demonstrations
· Robotics and Autonomous Systems
Robotics · 7 min text · 41 citations
-
Passive Obstacle Aware Control to Follow Desired Velocities
· IEEE Robotics and Automation Letters
Robotics · 8 min text · 40 citations
-
Modeling robot discrete movements with state-varying stiffness and damping: A framework for integrated motion generation and impedance control
· Robotics: Science and Systems (RSS 2014)
Robotics · 4 min text · 32 citations
-
Elastic Motion Policy: An Adaptive Dynamical System for Robust and Efficient One-Shot Imitation Learning
· IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025)
Robotics · 6 min text · 39 citations
-
Learning Contracting Vector Fields For Stable Imitation Learning
Preprint posted · arXiv
Robotics · 5 min text · 31 citations
-
Learning Variable Impedance Skills from Demonstrations with Passivity Guarantee
Preprint posted · arXiv
Robotics · 5 min text · 31 citations
-
VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon
· arXiv preprint
Robotics · 8 min text · 58 citations
-
Safety-Critical Event Triggered Control via Input-to-State Safe Barrier Functions
· IEEE Control Systems Letters
Robotics · 4 min text · 18 citations
-
Compatibility of Multiple Control Barrier Functions for Constrained Nonlinear Systems
· IEEE Conference on Decision and Control (CDC 2025)
Robotics · 4 min text · 32 citations
-
Composable Geometric Motion Policies using Multi-Task Pullback Bundle Dynamical Systems
· IEEE International Conference on Robotics and Automation (ICRA 2021)
Robotics · 7 min text · 28 citations
-
Stable Port-Hamiltonian Neural Networks
· NeurIPS 2025
Scientific machine learning · 5 min text · 57 citations
-
Globally Stable Neural Imitation Policies
· IEEE International Conference on Robotics and Automation (ICRA 2024)
Robotics · 5 min text · 29 citations
-
Neural Contractive Dynamical Systems
· ICLR 2024
Robotics · 6 min text · 44 citations
-
Contractive Dynamical Imitation Policies for Efficient Out-of-Sample Recovery
· ICLR 2025
Robotics · 8 min text · 50 citations
-
Stable Modular Control via Contraction Theory for Reinforcement Learning
· Learning for Dynamics and Control 2024
Robotics · 7 min text · 32 citations
-
Grounding Language Plans in Demonstrations Through Counterfactual Perturbations
· ICLR 2024
Robotics · 4 min text · 40 citations
-
End-to-End Stable Imitation Learning via Autonomous Neural Dynamic Policies
Presented · ICRA 2023 L3H2 Workshop
Robotics · 3 min text · 26 citations
-
Compositional Learning of Dynamical System Models Using Port-Hamiltonian Neural Networks
· Learning for Dynamics and Control 2023
Machine learning · 5 min text · 39 citations
-
Efficient Data Collection for Robotic Manipulation via Compositional Generalization
· Robotics: Science and Systems 2024
Robotics · 7 min text · 54 citations
-
Learning Modular Neural Network Policies for Multi-Task and Multi-Robot Transfer
· ICRA 2017
Robotics · 4 min text · 33 citations
-
Programmatically Grounded, Compositionally Generalizable Robotic Manipulation
· ICLR 2023
Robotics · 6 min text · 39 citations
-
Learning Task Constraints in Visual-Action Planning from Demonstrations
· RO-MAN 2021
Robotics · 4 min text · 22 citations
-
Sparse Control for Dynamic Movement Primitives
· IFAC-PapersOnLine
Robotics · 4 min text · 52 citations
-
Robot Control based on Motor Primitives — A Comparison of Two Approaches
· The International Journal of Robotics Research
Robotics · 9 min text · 119 citations
-
Scalable Learning of High-Dimensional Demonstrations with Composition of Linear Parameter Varying Dynamical Systems
· IEEE/RSJ International Conference on Intelligent Robots and Systems 2025
Robotics · 4 min text · 25 citations
-
Learning Compositional Behaviors from Demonstration and Language
· Conference on Robot Learning 2024
Robotics · 7 min text · 80 citations
-
Behavioral Cloning from Observation
· IJCAI 2018
Robotics · 4 min text · 23 citations
-
TACO: Learning Task Decomposition via Temporal Alignment for Control
· ICML 2018
Robotics · 5 min text · 27 citations
-
CompILE: Compositional Imitation Learning and Execution
· ICML 2019
Robotics · 8 min text · 55 citations
-
A Robust Layered Control System for a Mobile Robot
· MIT Artificial Intelligence Laboratory, AI Memo 864
Robotics · 5 min text · 15 citations
-
Learning Shared Safety Constraints from Multi-task Demonstrations
· NeurIPS 2023
Robotics · 6 min text · 47 citations
-
Incremental Composition of Learned Control Barrier Functions in Unknown Environments
· IEEE Conference on Decision and Control 2025
Robotics · 4 min text · 27 citations
-
Constrained Exploration and Recovery from Experience Shaping
· arXiv (preprint)
Robotics · 4 min text · 31 citations
-
Composable Energy Policies for Reactive Motion Generation and Reinforcement Learning
· Robotics: Science and Systems 2021
Robotics · 7 min text · 50 citations
-
RMP²: A Structured Composable Policy Class for Robot Learning
· Robotics: Science and Systems 2021
Robotics · 6 min text · 38 citations
-
RMPflow: A Computational Graph for Automatic Motion Policy Generation
· Algorithmic Foundations of Robotics XIII (WAFR 2018)
Robotics · 11 min text · 36 citations
-
Learning from Demonstration Using Products of Experts: Applications to Manipulation and Task Prioritization
· The International Journal of Robotics Research
Robotics · 9 min text · 69 citations
-
Learning Task Priorities from Demonstrations
· IEEE Transactions on Robotics
Robotics · 7 min text · 46 citations
-
Probabilistic Prioritization of Movement Primitives
· IEEE Robotics and Automation Letters
Robotics · 4 min text · 27 citations
-
Probabilistic Movement Primitives
· NIPS 2013
Robotics · 4 min text · 25 citations
-
Policy Architectures for Compositional Generalization in Control
· Reinforcement Learning Journal
Robotics · 6 min text · 39 citations
-
CCDP: Composition of Conditional Diffusion Policies with Guided Sampling
· IROS 2025
Robotics · 6 min text · 35 citations
-
Flexible Multitask Learning with Factorized Diffusion Policy
· IEEE Robotics and Automation Letters
Robotics · 6 min text · 45 citations
-
DexCompose: Reusing Dexterous Policies for Multi-Task Manipulation with a Single Hand
· arXiv (preprint)
Robotics · 6 min text · 43 citations
-
Cascaded Compositional Residual Learning for Complex Interactive Behaviors
· IEEE Robotics and Automation Letters
Robotics · 5 min text · 50 citations
-
MCP: Learning Composable Hierarchical Control with Multiplicative Compositional Policies
· NeurIPS 2019
Robotics · 7 min text · 42 citations
-
Critique of World Model
· arXiv (preprint)
World models · 8 min text · 72 citations
-
Learning Composable Skills by Discovering Spatial and Temporal Structure with Foundation Models
· ICRA 2026
Robotics · 9 min text · 123 citations
-
Understanding and Mitigating the Video-Action Generalization Gap via Temporal Ratio
Preprint posted · CoRL 2026 (accepted)
Robotics · 15 min text · 48 citations
-
Neuro-Symbolic Concepts
· Communications of the ACM
AI agents · 7 min text · 40 citations
-
WhatWorkedBench: Benchmarking Experimental Understanding in AI Agents
· arXiv (preprint)
Scientific discovery · 16 min text · 50 citations
-
Spatial-Interactor: Learning Spatial Reasoning through Interaction with the Observable Physical World
· arXiv (preprint)
Language models · 11 min text · 52 citations
-
HappyWorld-Bench
· arXiv (preprint)
World models · 28 min text · 125 citations
-
Beyond the Single Model: General Intelligence as a Property of Coupled Cognitive Architectures
Manuscript dated · PhilPapers (working paper)
AI agents · 5 min text · 19 citations
-
GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation
Preprint posted · arXiv (preprint)
World models · 15 min text · 76 citations
-
Lean Pool: An AI-Maintained Archive of Formalized Mathematics
· arXiv (preprint)
Scientific discovery · 10 min text · 231 citations
-
GAE: Learning a Geometry-Native Latent Space for 3D-Consistent World Generation
· arXiv (preprint)
World models · 10 min text · 62 citations
-
RULER: Instance-aware Rubric Rewards for SVG Generation
· arXiv (preprint)
Language models · 10 min text · 40 citations
-
The Tasteful Agent: Measuring and Improving Taste in Long-Horizon Tasks
Preprint posted · arXiv (preprint)
AI agents · 11 min text · 52 citations
-
Robot World Models Are Not Invariant to How the Actions Are Written
Preprint posted · arXiv (preprint)
World models · 12 min text · 16 citations
-
DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras
· Advances in Neural Information Processing Systems (NeurIPS)
Robotics · 5 min text · 61 citations
-
SuperVINS: A Real-Time Visual-Inertial SLAM Framework for Challenging Imaging Conditions
· IEEE Sensors Journal
Robotics · 4 min text · 24 citations
-
DynaVINS: A Visual-Inertial SLAM for Dynamic Environments
· IEEE Robotics and Automation Letters
Robotics · 6 min text · 26 citations
-
WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory
· arXiv (preprint)
World models · 5 min text · 105 citations
-
GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay
· arXiv (preprint)
AI agents · 6 min text · 50 citations
-
RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
· arXiv (preprint)
AI agents · 10 min text · 61 citations
-
Transferring the Intelligence of VLMs to Robotic Control
· arXiv (preprint)
Robotics · 5 min text · 55 citations
-
Neural basis of compositional control
· Nature
Neuroscience · 6 min text · 83 citations
-
A distributional code for value in dopamine-based reinforcement learning
· Nature
Neuroscience · 5 min text · 80 citations
-
Handwritten Digit classification with neural cultures is influenced by neural architecture, network dynamics, and decoding methods
· bioRxiv (preprint)
Neuroscience · 8 min text · 81 citations
-
Gated Memory Policy: In-Context Memorization and Adaptation
· arXiv (preprint)
Robotics · 7 min text · 49 citations
-
Jerk-limited Real-time Trajectory Generation with Arbitrary Target States
· Robotics: Science and Systems (RSS)
Robotics · 4 min text · 13 citations
-
Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents
· arXiv (preprint)
AI agents · 7 min text · 44 citations
-
JEPA-Anything: Learning Predictive Models across Different Worlds
· arXiv (preprint)
World models · 7 min text · 37 citations
-
Can MiniMax-H3 Reason About the Physical World? An Evaluation of Omni-Modal Generative Model
· arXiv (preprint)
World models · 4 min text · 48 citations
-
DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
· arXiv (preprint)
Language models · 10 min text · 99 citations
-
GeomVLA: Unifying Scene, Motion, and Action in 3D
Preprint posted · CoRL 2026 (accepted)
Robotics · 6 min text · 60 citations
-
CoMind: Understanding Collaborative Human Activity from Multiple Minds and Views
Preprint posted · ECCV 2026 (accepted)
Human collaboration · 6 min text · 97 citations
-
HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses
· arXiv (preprint)
Scientific discovery · 6 min text · 55 citations
-
DELE-w0.5: Inferring Action from Future Latent State for Robotic Manipulation
· arXiv (preprint)
Robotics · 5 min text · 55 citations
-
Dynamic Semantic Compression for Efficient Latent-Space Inference in Large Language Models
· arXiv (preprint)
Language models · 3 min text · 43 citations
-
Improving Policy Learning via Language-Guided State Representation in World Models
· IEEE Robotics and Automation Letters
Robotics · 4 min text · 47 citations
-
GaussVLA: Geometry-Aware Spatial Reasoning for Vision-Language-Action Model
Preprint posted · BMVC 2026 (accepted)
Robotics · 8 min text · 43 citations
No papers match your search.
Browse collections
Matching reading lists
Fei-Fei Li
2025–2026 papers on spatial intelligence, robotics, vision, and healthcare.
Hang Zhao
2025–2026 papers on robot learning, autonomy, and perception.
Jiayuan Mao
Research on concepts, reasoning, and robot learning, from 2017–2026.
Negar Mehr
2025–2026 papers on coordination, interaction, and control.
Roberto Martín-Martín
2025–2026 papers on learning practical robot skills.
Russ Tedrake
2025–2026 papers on planning, control, and learned robot behavior.
Wolfram Burgard
2025–2026 papers on mapping, navigation, and embodied learning.
Yann LeCun
Four sequences through visual representation learning, representation geometry, planning and control, and the design of predictive targets across languages, sound and scientific data. Reviewed through November 15, 2026.
CoRL 2026
Spotlight and oral sessions in program order, with links to all six keynote authors.
Spotlights
Orals
Anthropic’s interpretability works
A seven-paper sequence from transformer circuits to circuit tracing and biology.
Ilya’s reading collection
The complete 27-reading program: visual and sequence architectures, memory, information theory, intelligence and scalable training.
JEPA
45 papers by Yann LeCun and collaborators, organized into visual representations, geometry and theory, planning and control, and predictive targets across language, audio and scientific data.
Mathematical tools for language models
One complete guided path through 26 papers connecting geometry, attention, optimization, scaling, evaluation and distribution learning to language-model engineering.
Agents and reasoning
Reading sequences about agents and reasoning across authors and collections.
AI policy
Reading sequences about ai policy across authors and collections.
Foundations
Reading sequences about foundations across authors and collections.
Healthcare
Reading sequences about healthcare across authors and collections.
Interpretability
Reading sequences about interpretability across authors and collections.
Language models
Reading sequences about language models across authors and collections.
Manipulation
Reading sequences about manipulation across authors and collections.
Perception and mapping
Reading sequences about perception and mapping across authors and collections.
Planning and control
Reading sequences about planning and control across authors and collections.
Robot learning
Reading sequences about robot learning across authors and collections.
Vision and generation
Reading sequences about vision and generation across authors and collections.
World models
Reading sequences about world models across authors and collections.
No reading lists match these filters.
Anthropic's interpretability works
Follow Anthropic’s progression from small transformer circuits and induction heads to superposition, sparse features, and circuit tracing in larger language models.
- A Mathematical Framework for Transformer Circuits
- In-context Learning and Induction Heads
- Toy Models of Superposition
- Towards Monosemanticity: Decomposing Language Models With Dictionary Learning
- Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet
- Circuit Tracing: Revealing Computational Graphs in Language Models
- On the Biology of a Large Language Model
World models: physical consistency and robot control
Start with what a world model should preserve: geometry, remembered scenes and hidden objects. Examine editable scene programs and task-conditioned representations, then world–action models, timing, action-encoding failures and online correction. Video-generation benchmarks and robot-control experiments provide different kinds of evidence.
- Critique of World Model
- GAE: Learning a Geometry-Native Latent Space for 3D-Consistent World Generation
- WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory
- BWM-Forge: Agentic World Programming for Embodied Intelligence
- Training Object Permanence in World Models
- Can MiniMax-H3 Reason About the Physical World? An Evaluation of Omni-Modal Generative Model
- HappyWorld-Bench
- Improving Policy Learning via Language-Guided State Representation in World Models
- DELE-w0.5: Inferring Action from Future Latent State for Robotic Manipulation
- GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation
- LeWAM: End-to-End World and Action Modeling with JEPAs
- GreenWAM: A World Foundation Model for Robotics
- Understanding and Mitigating the Video-Action Generalization Gap via Temporal Ratio
- Robot World Models Are Not Invariant to How the Actions Are Written
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon
Reusable robot skills: discovery, composition and transfer
Begin with modular policies and learning skill boundaries from demonstrations. Progress to composable control primitives, grounded programs, language and geometric skill effects, then compare residual and diffusion composition before dexterous reuse. Read how each method defines a skill and when its composition transfers.
- Learning Modular Neural Network Policies for Multi-Task and Multi-Robot Transfer
- TACO: Learning Task Decomposition via Temporal Alignment for Control
- CompILE: Compositional Imitation Learning and Execution
- MCP: Learning Composable Hierarchical Control with Multiplicative Compositional Policies
- Programmatically Grounded, Compositionally Generalizable Robotic Manipulation
- Grounding Language Plans in Demonstrations Through Counterfactual Perturbations
- Learning Compositional Behaviors from Demonstration and Language
- Learning Composable Skills by Discovering Spatial and Temporal Structure with Foundation Models
- Cascaded Compositional Residual Learning for Complex Interactive Behaviors
- Flexible Multitask Learning with Factorized Diffusion Policy
- CCDP: Composition of Conditional Diffusion Policies with Guided Sampling
- DexCompose: Reusing Dexterous Policies for Multi-Task Manipulation with a Single Hand
Mechanistic interpretability: features, circuits and causal tests
Begin with the transformer path framework and superposition, then build and evaluate sparse feature dictionaries. Use those features to trace computations, connect attention to attribution graphs, and finish with faithfulness counterexamples and counterfactual tests. Keep reconstruction quality, useful features and a correct causal explanation distinct.
- A Mathematical Framework for Transformer Circuits
- Toy Models of Superposition
- Towards Monosemanticity: Decomposing Language Models With Dictionary Learning
- Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet
- Using Dictionary Learning Features as Classifiers
- Evaluating Sparse Autoencoders on Targeted Concept Erasure Tasks
- SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
- Circuit Tracing: Revealing Computational Graphs in Language Models
- On the Biology of a Large Language Model
- Tracing Attention Computation Through Feature Interactions
- A Toy Model of Mechanistic (Un)Faithfulness
- Would this change your answer? Evaluating Explanations of LLM Behavior In The Wild with Counterfactual Experiments
Model behavior: steering, introspection and auditing
Study how activation directions change refusal, traits and conversational behavior, then use internal signals to investigate concealed objectives. Progress from probes and blinded audits to learned behavior descriptions, activation explanations and auditing agents. Interventions show controllability; audits test what can be uncovered under their stated threat models.
- Steering Llama 2 via Contrastive Activation Addition
- Refusal in Language Models Is Mediated by a Single Direction
- Evaluating feature steering: A case study in mitigating social biases
- persona vectors: monitoring and controlling character traits in language models
- Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning
- The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
- Simple probes can catch sleeper agents
- Auditing Language Models for Hidden Objectives
- Introspection Adapters: Training LLMs to Report Their Learned Behaviors
- Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
- Building and evaluating alignment auditing agents
- AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors
Visual representations for reasoning and transfer
How does representation learning help with unfamiliar reasoning problems? Compare a neuro-symbolic concept framework, natural-image pretraining for ARC, and interaction-trained spatial reasoning; these test different kinds of transfer.
Stable imitation and recovery
How can an imitation policy return to useful motion after a disturbance? Trace contraction and Lyapunov methods from learned vector fields to neural policies, unfamiliar starting states, and adaptation to moving objects.
- Learning Contracting Vector Fields For Stable Imitation Learning
- End-to-End Stable Imitation Learning via Autonomous Neural Dynamic Policies
- Neural Contractive Dynamical Systems
- Globally Stable Neural Imitation Policies
- Contractive Dynamical Imitation Policies for Efficient Out-of-Sample Recovery
- Elastic Motion Policy: An Adaptive Dynamical System for Robust and Efficient One-Shot Imitation Learning
Combining task-space motion objectives
How can several task-space preferences determine one robot command? Start with probabilistic movement primitives and learned priorities, then compare products of experts with geometric pullbacks and differentiable policy composition.
- Probabilistic Movement Primitives
- Probabilistic Prioritization of Movement Primitives
- Learning from Demonstration Using Products of Experts: Applications to Manipulation and Task Prioritization
- RMPflow: A Computational Graph for Automatic Motion Policy Generation
- Composable Geometric Motion Policies using Multi-Task Pullback Bundle Dynamical Systems
- RMP²: A Structured Composable Policy Class for Robot Learning
Improving agents beyond model weights
How do skills, tools, and memory improve agents without changing model weights? Study skill evolution, harness optimization, and visual inspection before examining decision quality and experimental understanding.
- Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents
- RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
- VISTA: A Visual Harness for Reasoning in an Interactive World
- The Tasteful Agent: Measuring and Improving Taste in Long-Horizon Tasks
- WhatWorkedBench: Benchmarking Experimental Understanding in AI Agents
Ilya’s reading list — the complete 27-reading program
All 27 documented entries in the circulated selection attributed to Ilya Sutskever, arranged as one guided program. Begin with visual features and residual networks, move through sequence modeling, attention, structured reasoning and memory, then study compression, complexity and definitions of intelligence. Finish with distributed training and scaling laws. This thematic reading order is editorial; the catalog separately preserves the circulated source order.
- CS231n: Convolutional Neural Networks for Visual Recognition
- ImageNet Classification with Deep Convolutional Neural Networks
- Deep Residual Learning for Image Recognition
- Identity Mappings in Deep Residual Networks
- Multi-Scale Context Aggregation by Dilated Convolutions
- The Unreasonable Effectiveness of Recurrent Neural Networks
- Understanding LSTM Networks
- Recurrent Neural Network Regularization
- Neural Machine Translation by Jointly Learning to Align and Translate
- Attention Is All You Need
- The Annotated Transformer
- Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
- Pointer Networks
- Order Matters: Sequence to Sequence for Sets
- Neural Turing Machines
- Neural Message Passing for Quantum Chemistry
- A simple neural network module for relational reasoning
- Relational recurrent neural networks
- Keeping Neural Networks Simple by Minimizing the Description Length of the Weights
- Variational Lossy Autoencoder
- A Tutorial Introduction to the Minimum Description Length Principle
- The First Law of Complexodynamics
- Quantifying the Rise and Fall of Complexity in Closed Systems: The Coffee Automaton
- Kolmogorov Complexity and Algorithmic Randomness — Algorithmic Statistics and Appendices
- Machine Super Intelligence
- GPipe: Easy Scaling with Micro-Batch Pipeline Parallelism
- Scaling Laws for Neural Language Models
CoRL 2026 Spotlight S1 · Generalist policies, memory and transfer
How does a generalist policy know which skill to use? Read experiments on new instruction combinations, transfer between robot bodies, memory, and conflicts between what the robot sees and what it is told.
- Diagnosing Compositional Generalization in Sequential Robot Tasks
- EmbodiSteer: Steering Embodiment-Agnostic Visuomotor Policies with Joint-Space Guidance for Zero-Shot Cross-Embodiment Deployment
- Learning What to Say to Your VLA: Mostly Harmless Vision Language Action Model Steering
- MEM: Multi-Scale Embodied Memory for Vision Language Action Models
- MolmoAct2: Action Reasoning Models for Real-World Deployment
- PRTS: A Primitive Reasoning and Tasking System via Contrastive Representations
- Semantic Anchoring for Robotic Action Representations
- When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs
- ZETA: A Controlled Study of Zero-Shot Cross-Embodiment VLA Transfer for Tabletop Manipulation
CoRL 2026 Oral O1 · Generalist policies and context scaling
Compare three ways to make one policy useful in a new setting. Being-H0.5 learns from human interaction, RoboTTT extends the robot’s history, and π₀.₇ lets a user steer behavior through context.
CoRL 2026 Oral O2 · World models and temporal representations
What does a robot need to remember or predict before it moves? These papers use a persistent 3D scene, a short motion history, and joint prediction of video and actions.
CoRL 2026 Oral O3 · Data, simulation and evaluation
Follow the evidence from training data to evaluation. The papers measure demonstration diversity, test how far simulation transfers to hardware, and audit the shortcuts a manipulation benchmark can reward.
CoRL 2026 Spotlight S2 · Efficient execution, world models and learning
Compare changes to policy size, model-based training, and online inference. The last two papers ask when a controller should change actions and how the simulator used for training can affect the resulting policy.
- BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation
- Chunk-wise Model-based Policy Optimization
- Diffusion ReRoll: Revisable Denoising for Robotic Sequential Prediction
- EagleVLA: Towards Onboard Real-Time Robot Control via Foresight-Aligned Asynchronous Inference
- FedGuide: Diffusion Prior Alignment and Value Baseline Guidance for Heterogeneous Federated Reinforcement Learning
- MoRE: Mixture-of-Experts for Multi-Physics Robotic World Models
- REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models
- TARC: Time-Adaptive Robotic Control
- Targeting World Models to Compromise Robot Learning Pipelines
CoRL 2026 Spotlight S3 · Human learning, dexterity and cooperation
Trace how a demonstration becomes a dexterous or cooperative skill. The papers use paired hands, operator corrections, first-person human video, gaze, partner preferences, and touch.
- BiDexGrasp: Coordinated Bimanual Dexterous Grasps across Object Geometries and Sizes
- Efficient and Reliable Teleoperation through Real-to-Sim-to-Real Shared Autonomy
- Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data
- FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space
- KPGrasp: Scalable Keypoint Flow Matching for Dexterous Grasp Generation
- Learning Human Intention for Robotic Manipulation
- Leave It Better Than Found: Value-Guided Cooperation in Sequential Tasks
- Robots that Collaborate: Sequential Asymmetric Imitation for Learning Coupled Robot Policies
- WM-Craftnet: World Synesthesia Model for Robust and Generalizable Dexterous In-Hand Manipulation
CoRL 2026 Oral O4 · Reinforcement learning and motion reasoning
Compare a motion forecaster constrained by traffic rules with two approaches to reinforcement learning. One trains flow-based controllers, while the other adapts a pretrained robot policy through a compact action representation.
CoRL 2026 Oral O5 · Dexterous learning, grasping and assembly
Start with learning hand motions from human video, then study how a robot chooses a grasp for a named purpose and how pretraining through play transfers to precise assembly.
CoRL 2026 Oral O6 · Force, tactile control and manipulation planning
Force estimation, physics-based search, and tactile feedback address different moments in manipulation. Read how the robot infers a load, plans a sequence of contacts, and corrects its actions at insertion time.
CoRL 2026 Oral O7 · Humanoid data, control and skills
Compare the data and controllers behind three whole-body skills. The papers generate interactions for mobile manipulation, train a soccer kick, and use a differentiable simulator to learn motion tracking.
CoRL 2026 Spotlight S4 · Touch, force and contact-rich manipulation
When does contact information change a robot’s action? Compare how policies sample contact states, estimate constraints, sense underwater contact, and turn visual or force measurements into timely corrections.
- Combining Constrained Sampling and Reinforcement Learning for Manipulation
- Constraint-Decomposed Impedance Control from Generative Policy for Articulated Object Manipulation
- DAVATaR: Depth-Aware Visually Augmented Tactile Learning Across the Air-Water Interface
- Demystifying When and Why VLAs Fail in Contact-Rich Tasks and How to Fix Them
- DeTac: Dual-branch Decoupling of Texture and Force Signals in Vision-Based Tactile Sensors
- FingerEye: Learning Dexterous Manipulation with Continuous Vision-Tactile Sensing
- Never Too Late for Force: Accelerating VLA Post-Training with Reactive Force Injection
- TAM: Torque Adaptation Module for Robust Motion Transfer in Manipulation
- Visible Touch: Rendering Contact for Visuomotor Policies
CoRL 2026 Spotlight S5 · Deformable manipulation, geometry and morphology
Compare what a robot must represent to manipulate an articulated scene, a cloth, or a wire. The session also studies depth uncertainty, robot reachability, and how changing the body changes the task.
- Articulated 3D Scene Graphs for Open-World Mobile Manipulation
- Enabling Robust Cloth Manipulation via Inference-Time Simulator-in-the-Loop Refinement
- Learning Physics-Guided Residual Dynamics for Deformable Object Simulation
- Noise-Invariant Spatial Encoding for 3D Manipulation Policies under Depth Corruption
- RAM: Reachability Across Morphologies
- Robustness without Wrinkles: Parallel Simulation and Robust MPC for Certified Deformable Manipulation
- SERF: Spatiotemporal Environment and Robot Feature Map for Long-Horizon Mobile Manipulation
- Shape Your Body: Value Gradients for Multi-Embodiment Robot Design
- WireCraft: A Simulation Benchmark for Industrial DLO Manipulation
CoRL 2026 Spotlight S6 · Optimization, agentic planning and evaluation
How can a robot choose and check a plan? Start with evaluations of compositional skills and user intent, then compare uncertainty estimates, constrained plan generation, exploratory learning, and numerical solvers.
- ATOM-Bench: A Real-World Benchmark for Atomic Skills and Compositional Generalization in Manipulation Policies
- Bayesian Active Learning for Intent Disambiguation in Interactive Robot Planning
- Control-Ready Uncertainty for Trajectory Diffusion
- Emergent Control via Prior-Based Free Energy Minimization
- GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks
- Meta-Ctrl: Guaranteed Plan Generation by Decoupling Syntactic and Semantic Constraints
- Playful Agentic Robot Learning
- RoboKit: Accelerating Parallel Differentiable Optimization for Robot Learning
- Scalable Deep Unfolding of Conic Optimizers
CoRL 2026 Spotlight S7 · Humanoid locomotion and whole-body control
Read from whole-body motion tracking and cluttered navigation through simulation, ladder climbing, and learning from demonstrations. The final papers look at feasible reference motions, safety corrections, and pressure measured underfoot.
- CMP: Robust Whole-Body Tracking for Loco-Manipulation via Competence Manifold Projection
- Humanoid Navigation in Cluttered Environments with a Whole-Body Vision-Language-Action Model
- Kamino: GPU-based Simulation of Rigid Multi-Body Physics with Kinematic Loops
- LadderMan: Learning Humanoid Perceptive Ladder Climbing
- Learning Loco-Manipulation From SMPC Demonstrations With Sparse Offline-to-Online RL
- OpenHLM: An Empirical Recipe for Whole-Body Humanoid Loco-Manipulation
- PHUMA: Physically Reliable Humanoid Locomotion Dataset
- ResSafe: Learning Safety Filtering with Residual Reinforcement Learning for Humanoids
- TactileStep: Sole Tactile Learning for Regulating Foot–Terrain Interaction in Humanoid Locomotion
CoRL 2026 Spotlight S8 · Perception, navigation and autonomous systems
What must an autonomous system know to decide where to go? These studies cover 3D reconstruction, coordination, finding a specific target, estimating motion, waiting for obstacles, docking, and evaluating sensors and planners.
- Anchor3R: Streaming 3D Reconstruction with Transient Anchors for Long-Horizon Visual Mapping
- Collaborative Planning for UAV-UGV Teams under Operational Constraints via Masked Error-bounded Neuro-Symbolic Diffusion
- From Region Arrival to Instance-Level Grounding in Vision-and-Language Navigation
- MosaicIMU: Composing Carrier Experts for Generalizable Neural Inertial Odometry
- OSCAR: Obstacle Survival Curves for Adaptive Robot Navigation
- SCoCaT: Success Conditioned Constrained Reinforcement Learning for Spacecraft Docking
- SidewalkBench: Benchmarking Visual Navigation on Urban Sidewalks
- TRIAD Planner: Three-Level Robust Imitation Learning for Autonomous Driving
- VisionSIM: Beyond Idealized Sensors — A Multi-Modal Simulator for Robot Perception
Hang Zhao · Robot foundation models: data, actions and world models
Follow the full learning pipeline from robot and human data through action tokenizers, autoregressive control and predictive world-action models. Compare what is learned during pretraining, what is imagined during execution, and what transfers between embodiments.
- RoboCOIN: An Open-Sourced Bimanual Robotic Data Collection for Integrated Manipulation
- Galaxea Open-World Dataset and G0 Dual-System VLA Model
- Dexora: Open-source VLA for High-DoF Bimanual Dexterity
- UniDex: A Robot Foundation Suite for Universal Dexterous Hand Control from Egocentric Human Videos
- ActionCodec: What Makes for Good Action Tokenizers
- FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via Neural Action Tokenization
- Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching
- Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models
- G0.5: One Autoregressive Stream for Robot Reasoning and Action
- ORV: 4D Occupancy-centric Robot Video Generation
- Orca: The World is in Your Mind
- OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining
- Representation World Model: Learning States, Transition and Executable Plans in Representation
- Fast-WAM: Do World Action Models Need Test-time Future Imagination?
- SkelWAM: A Skeleton-Guided World-Action Model for Zero-Shot Cross-Embodiment Manipulation
- WB-WAM: Heterogeneous Body-Hand Pre-training for Humanoid Loco-Manipulation
Hang Zhao · Manipulation policies: geometry, adaptation and dexterity
Compare data augmentation, diffusion and flow-based control, depth and occupancy representations, then contact-preserving retargeting and dexterous interaction. The progression connects policy inputs and action generators to adaptation on physical robots.
- RoboEngine: Plug-and-Play Robot Data Augmentation with Semantic Robot Segmentation and Background Generation
- Conditioning Matters: Training Diffusion Policies is Faster Than You Think
- Generative Control as Optimization: Time Unconditional Flow Matching for Adaptive and Robust Robotic Control
- DepthVLA: Enhancing Vision-Language-Action Models with Depth-Aware Spatial Reasoning
- OccVLA: Vision-Language-Action Model with Implicit 3D Occupancy Supervision
- VolumeDP: Modeling Volumetric Representation for Manipulation Policy Learning
- TopoRetarget: Interaction-Preserving Retargeting for Dexterous Manipulation
- DexNinja: Learning Robust Dexterous Cutting Policy with a Real-to-Sim-to-Real Data Engine
- EqC: Equilibrium Energy Control for Introspective Lifelong Robot Learning
Hang Zhao · Spatial reconstruction, occupancy and persistent mapping
Trace representations of space from multi-view images and occupancy to dynamic reconstruction, streaming SLAM and persistent world models. The final remote-sensing study extends the discussion to reconstructing spatial evidence over much longer time scales.
- BEV-VAE: Multi-view Image Generation with Spatial Consistency for Autonomous Driving
- Diffusion-Based Generative Models for 3D Occupancy Prediction in Autonomous Driving
- GS-Occ3D: Scaling Vision-Only Occupancy Reconstruction with Gaussian Splatting
- TrackOcc: Camera-Based 4D Panoptic Occupancy Tracking
- LONG3R: Long Sequence Streaming 3D Reconstruction
- Complet4R: Geometric Complete 4D Reconstruction
- SLAM-Former: Putting SLAM Into One Transformer
- SLAMFormer-∞: Infinite SLAM Transformer for Unbounded Frontend and Backend Processing
- World SLAM Model: Joint World Modeling for SLAM and Navigation
- Global reconstruction of three decades of fine-grained nighttime light data with analysis of large-scale infrastructure and landmarks
Hang Zhao · Autonomous driving: perception, planning and evaluation
Study lane topology and mapping uncertainty before trajectory choices, driver preferences and language-conditioned action plans. Then follow the data and simulation pipeline through generated corner cases, safety metrics and arguments about evidence for reliable driving.
- Chameleon: Fast-Slow Neuro-Symbolic Lane Topology Extraction
- Reusing Attention for One-stage Lane Topology Understanding
- Delving into Mapping Uncertainty for Mapless Trajectory Prediction
- Generalizing Motion Planners with Mixture of Experts for Autonomous Driving
- Learning Personalized Driving Styles via Reinforcement Learning from Human Feedback
- DAP: A Discrete-token Autoregressive Planner for Autonomous Driving
- Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving
- DriveMA: Driving Vision-Language-Action Models with verifiable Meta-Actions
- DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking
- Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models
- CRUISE: Cooperative Reconstruction and Editing in V2X Scenarios using Gaussian Splatting
- Gen-NCAP: A Generative Simulator for Corner Case Benchmarking in End-to-End Autonomous Driving
- Challenger: Affordable Adversarial Driving Video Generation for Safety Testing
- Agent-driven Long-tail Simulation for Autonomous Driving
- Is Your Trajectory Displacement Safe in Long-tail?
- Position: Prospective of Autonomous Driving - Multimodal LLMs, World Models, Embodied Intelligence, AI Alignment, and Mamba
- Evolution path of multimodal collaborative optimization and training techniques for end-to-end autonomous driving systems
- The Second Half of Autonomous Driving: Debating the Road to N-Nines Reliability
Hang Zhao · Speech, spatial hearing and expressive robots
Connect speech translation and acoustic sensing with the mechanisms that make a robot communicate physically. Read from speech units and spatial sound to adaptive sensing, expressive actuation and automated facial-mechanism design.
- When End-to-End is Overkill: Rethinking Cascaded Speech-to-Text Translation
- A Unit-based System and Dataset for Expressive Direct Speech-to-Speech Translation
- Supervising Sound Localization by In-the-wild Egomotion
- SPACE: Speaker Adaptation for Acoustic Eavesdropping Using mmWave Radio Signals
- Morpheus: A Neural-driven Animatronic Face with Hybrid Actuation and Diverse Emotion Control
- Automated Synthesis of Facial Mechanisms for Conversational Animatronic Robots
Hang Zhao · Language-model context, verification and circuits
Compare three ways to analyze or improve language-model computation: the behavior of longer context, reinforcement learning for formal software verification, and decomposing weights to extract circuits. These collaborations address model structure and evidence rather than robot embodiment.
Negar Mehr · Multi-agent learning, games and coordination
Start with strategic interaction and models of other agents, then learn from demonstrations, credit assignment and coaching. The sequence connects game-theoretic assumptions and information limits to partner diversity, decentralized policies and coordination at execution time.
- Strategic Decision-Making in Multiagent Domains: A Weighted Constrained Potential Dynamic Game Approach
- MultiNash-PF: A Particle Filtering Approach for Computing Multiple Local Generalized Nash Equilibria in Trajectory Games
- Active Inverse Learning in Stackelberg Trajectory Games
- When Should a Leader Act Suboptimally? The Role of Inferability in Repeated Stackelberg Games
- Multi-Agent Inverse Reinforcement Learning in Real World Unstructured Pedestrian Crowds
- Understanding and Imitating Human-Robot Motion with Restricted Visual Fields
- Leveraging Large Language Models for Effective and Explainable Multi-Agent Credit Assignment
- CRAFT: Coaching Reinforcement Learning Autonomously using Foundation Models for Multi-Robot Coordination Tasks
- MIMIC-D: Multi-modal Imitation for MultI-agent Coordination with Decentralized Diffusion Policies
- Matching Multiple Experts: On the Exploitability of Multi-Agent Imitation Learning
- PopGen: Generating Semantically Diverse Populations for Zero-Shot Coordination
- Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand
- Congestion Games with Heterogeneous Valuations: An Optimal Transport Approach
- A Robust Approach to Belief Mismatch in Non-cooperative Dynamic Games
Negar Mehr · Adaptive control: constraints, curricula and computation
Connect admissible trajectories and action safeguards to how controllers are learned, retained and generated. Finish with batched nonlinear optimization and derivatives, which make these control computations practical at scale.
- DDAT: Diffusion Policies Enforcing Dynamically Admissible Robot Trajectories
- Learning Control Policies to Provably Satisfy Hard Affine Constraints for Black-Box Hybrid Dynamical Systems
- X-Safe: Embodiment-Agnostic Safeguarding for Robot Arm Manipulation Policies via Action Masking
- CurricuLLM: Automatic Task Curricula Design for Learning Complex Robot Skills Using Large Language Models
- HuGo: LLMs as Whole-Body Policy Code Designers for Humanoid Loco-Manipulation
- SAMBAR: Selective Anchoring via Method of Multipliers for Balanced Knowledge Acquisition and Retention in Vision-Language-Action Models
- Scaling Nonlinear Optimization: Many Problems One GPU
- CusADi2: Latency-Optimized Batch Computation of Robot Kinematics, Jacobians, and Hessians on the GPU
Negar Mehr · Distributed and continual neural mapping
Read the progression from asynchronous multi-robot maps to uncertainty-weighted communication and temporal consensus. The shared question is how a map stays consistent as robots receive different observations or must retain earlier knowledge.
Negar Mehr · Space robotics: risk, collection and servicing
Compare three mission-level applications: risk-sensitive orbital collision avoidance, robotic material collection and a servicing architecture for a space observatory. The small group is kept distinct because its operating constraints and evidence differ from terrestrial learning benchmarks.
Roberto Martín-Martín · Learning robot policies from data and feedback
Begin with a survey of physical reinforcement learning, then trace demonstration selection, simulated augmentation and iterative data collection. Corrections, safe transfer, reinforcement-learning fine-tuning and simulation-based experimentation show how policies continue improving after their initial data collection.
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes
- COLLAGE: Adaptive Fusion-based Retrieval for Augmented Policy Learning
- DataMIL: Selecting Data for Robot Imitation Learning with Datamodels
- MoMaGen: Generating Demonstrations under Soft and Hard Constraints for Multi-Step Bimanual Mobile Manipulation
- MOMASPLAT: Dynamic Novel-View Demonstration Augmentation for Mobile Manipulation
- GUILD: Guided Iterative Data Collection for Imitation Learning
- SafeMimic: Towards Safe and Autonomous Human-to-Robot Imitation for Mobile Manipulation
- ReDex: Repairing Sim-to-Real Dexterous Policies by Finger-Level Compliant Interaction
- OopsieRecover: Learning Transferable VLA Corrections from Semantic Interventions
- FLaRe: Achieving Masterful and Adaptive Robot Policies with Large-Scale Reinforcement Learning Fine-Tuning
- SLAC: Safe and Efficient Real-Robot Reinforcement Learning via Unsupervised Simulation Pre-Training
- Simple Recipe Works: Vision-Language-Action Models are Natural Continual Learners with Reinforcement Learning
- SimEX: Simulation-Integrated Robotics AutoResearch
Roberto Martín-Martín · Memory, world models and embodied reasoning
Compare mechanisms for carrying observations and demonstrations across time, then use them for search, questioning and partially observed planning. Learned dynamics, geometric features and integrated reasoning-and-action systems connect persistent state to long-horizon control.
- Scaling Short-Term Memory of Visuomotor Policies for Long-Horizon Tasks
- Memory Retrieval in Visuomotor Policies for Long-Horizon Robot Control
- Beyond Episodic Evaluation: Memory Architectural Bottlenecks in Sequential Embodied Question Answering
- RoboSSM: Scalable In-context Imitation Learning via State-Space Models
- MimicDroid: In-Context Learning for Humanoid Robot Manipulation from Human Play Videos
- Searching in Space and Time: Unified Memory-Action Loops for Open-World Object Retrieval
- Large-Language-Model-Guided State Estimation for Partially Observable Task and Motion Planning
- Factored Latent Action World Models
- EgoExo-WM: Unlocking Exo Video for Ego World Models
- Understanding the Impact of Geometric Foundation Models on Vision-Language-Action Models
- BUMBLE: Unifying Reasoning and Acting with Vision-Language Models for Building-wide Mobile Manipulation
Roberto Martín-Martín · Dexterity, contact and physical interaction
Start with a perspective on learning through contact, then compare compliant hardware, tactile and acoustic sensing, tool use and object-state changes. End with deformable dynamics, compositional dexterity and force-aware humanoid interaction.
- Robots Learning Through Physical Interactive Intelligence
- BiFlex: A Passive Bimodal Stiffness Flexible Wrist for Manipulation in Unstructured Environments
- DAVATaR: Depth-Aware Visually Augmented Tactile Learning Across the Air-Water Interface
- CAVER: Curious Audiovisual Exploring Robot
- GeT-USE: Learning Generalized Tool Usage for Bimanual Mobile Manipulation via Simulated Embodiment Extensions
- Mash, Spread, Slice! Learning to Manipulate Object States via Visual Spatial Progress
- Rapid Adaptation of Particle Dynamics for Generalized Deformable Object Mobile Manipulation
- CoDex: Learning Compositional Dexterous Functional Manipulation without Demonstrations
- VOFA: Visual Object Goal Pushing with Force-Adaptive Control for Humanoids
Roberto Martín-Martín · Evaluation, safety and human collaboration
Read how generalist robots are evaluated in physical and simulated settings before examining whose goals and preferences their actions should serve. Assistive intent, mixed-initiative dialogue and accessible communication connect benchmark performance to collaboration with people.
- Principles and Guidelines for Evaluating Social Robot Navigation Algorithms
- RoboArena: Distributed Real-World Evaluation of Generalist Robot Policies
- OopsieVerse: A Safety Benchmark with Damage-Aware Simulation for Robot Manipulation
- Why Automate This? Exploring the Connection between Time Use, Well-being and Robot Automation Across Social Groups
- CASPER: Inferring Diverse Intents for Assistive Teleoperation with Vision Language Models
- Mixed-Initiative Dialog for Human-Robot Collaborative Manipulation
- Designing Accessible Robot Communication for Blind People
Jiayuan Mao · Neuro-symbolic concepts and grounded language
Read the neuro-symbolic overview, then connect visual concepts and relations with structured language, syntax and learned lexicons. Work on writing systems extends the same search for reusable structure; the dissertation closes the path with a broader synthesis of learning, reasoning and planning.
- Neuro-Symbolic Concepts
- The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervision
- Visual Concept-Metaconcept Learning
- FALCON: Fast Visual Concept Learning by Integrating Images, Linguistic Descriptions, and Conceptual Relations
- NS3D: Neuro-Symbolic Grounding of 3D Objects and Relations
- What's Left? Concept Grounding with Logic-Enhanced Foundation Models
- Neural Phrase-to-Phrase Machine Translation
- Learning Visually-Grounded Semantics from Contrastive Adversarial Samples
- Unified Visual-Semantic Embeddings: Bridging Vision and Language With Structured Meaning Representations
- Visually Grounded Neural Syntax Acquisition
- Language-Mediated, Object-Centric Representation Learning
- Grammar-Based Grounded Lexicon Learning
- Finding structure in logographic writing with library learning
- Finding structure in logographic writing with library learning II: Grapheme, sound, and meaning systematicity
- Learning, Reasoning, and Planning with Neuro-Symbolic Concepts
Jiayuan Mao · Visual, relational and generative reasoning
Start with physical events and compositional visual reasoning, then study logical, relational and hypergraph computation. Structure inference and iterative energy-based reasoning connect these architectures to generative synthesis, with attention theory and problem-solving evaluations clarifying their limits.
- Grounding Physical Concepts of Objects and Events Through Dynamic Visual Reasoning
- CLEVRER-Humans: Describing Physical and Causal Events the Human Way
- Object-Centric Diagnosis of Visual Reasoning
- DisCo: Improving Compositional Generalization in Visual Reasoning through Distribution Coverage
- Neural Logic Machines
- On the Expressiveness and Generalization of Hypergraph Neural Networks
- Sparse and Local Networks for Hypergraph Reasoning
- Temporal and Object Quantification Networks
- What Planning Problems Can A Relational Neural Network Solve?
- Neurally-Guided Structure Inference
- Learning Iterative Reasoning through Energy Diffusion
- What Makes a Maze Look Like a Maze?
- Composition-Grounded Data Synthesis for Visual Reasoning
- Generating Fine Details of Entity Interactions
- V-GRPO: Online Reinforcement Learning for Denoising Generative Models Is Easier than You Think
- Learning Linear Attention in Polynomial Time
- Creativity or Brute Force? Using Brainteasers as a Window into the Problem-Solving Abilities of Large Language Models
Jiayuan Mao · Program induction for scenes, motion and assembly
Move from detection and localization to programs that describe images, planar scenes and human motion. Assembly manuals then show how a structured visual description can become an executable sequence of actions.
- What Can Help Pedestrian Detection?
- Acquisition of Localization Confidence for Accurate Object Detection
- Program-Guided Image Manipulators
- Perspective Plane Program Induction From a Single Image
- Multi-Plane Program Induction with 3D Box Priors
- Hierarchical Motion Understanding via Motion Programs
- Programmatic Concept Learning for Human Motion Description and Synthesis
- Translating a Visual LEGO Manual to a Machine-Executable Plan
- IKEA-Manual: Seeing Shape Assembly Step by Step
Jiayuan Mao · Structured planning and composable robot skills
Connect learned planning abstractions and hybrid symbolic-continuous representations to compositional constraint solving and action planning. Then ground those plans in reusable agents, part-based skills, demonstrations and language, comparing what must be specified with what can be learned.
- PDSketch: Integrated Domain Programming, Learning, and Planning
- Learning Adaptive Planning Representations with Natural Language Guidance
- Learning Planning Abstractions from Language
- Hybrid Declarative-Imperative Representations for Hybrid Discrete-Continuous Decision-Making
- Compositional Diffusion-Based Continuous Constraint Solvers
- “Set It Up!”: Functional Object Arrangement with Compositional Generative Models
- SIMPACT: Simulation-Enabled Action Planning using Vision-Language Models
- Discovering Hybrid World Representations with Co-Evolving Foundation Models
- Universal Agent for Disentangling Environments and Tasks
- Programmatically Grounded, Compositionally Generalizable Robotic Manipulation
- Composable Part-Based Manipulation
- Learning Compositional Behaviors from Demonstration and Language
- Learning Composable Skills by Discovering Spatial and Temporal Structure with Foundation Models
- KALM: Keypoint Abstraction Using Large Models for Object-Relative Imitation Learning
- Grounding Language Plans in Demonstrations Through Counterfactual Perturbations
Jiayuan Mao · Learning manipulation from video, contact and force
Follow actionless video and visual replanning into the gap between predicted videos and executable control. Reusable contact strategies, multimodal policies, muscle-derived force and grasp synthesis then compare the physical information needed for manipulation.
- Learning to Act from Actionless Videos through Dense Correspondences
- Implicit State Estimation via Video Replanning
- Understanding and Mitigating the Video-Action Generalization Gap via Temporal Ratio
- Learning Reusable Manipulation Strategies
- One-Shot Manipulation Strategy Learning by Making Contact Analogies
- Multi-Modal Manipulation via Multi-Modal Policy Consensus
- ForceBand: Learning Forceful Manipulation with sEMG
- Adversarial Game-Theoretic Algorithm for Dexterous Grasp Synthesis
Jiayuan Mao · Language-guided agents, intent and interaction
Begin with communication, demonstrated subgoals and inferred human intent. Embodied interfaces, action-based questioning, workflow memory and evaluation harnesses show how a language-guided agent organizes decisions beyond an individual manipulation policy.
- HandMeThat: Human-Robot Communication in Physical and Social Environments
- Learning Rational Subgoals from Demonstrations and Instructions
- Infer Human's Intentions Before Following Natural Language Instructions
- Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making
- ActionEQA: Action Interface for Embodied Question Answering
- Agent Workflow Memory
- Guava: An Effective and Universal Harness for Embodied Manipulation
Wolfram Burgard · Robotics foundations, systems and evaluation
Begin with robotics foundations and the AI roadmap, then inspect the control stack, policy-serving infrastructure and reproducible bimanual benchmark needed to turn learning methods into usable robot systems.
Wolfram Burgard · Spatial perception, mapping and localization
Study graph-based SLAM and registration, then representations that support relocalization, persistent world memory and semantic maps. Label-efficient segmentation, articulated-object estimation and open-vocabulary scene graphs add the perceptual structure those maps need.
- An Updated Tutorial on Graph-Based SLAM
- End-to-end 2D-3D Registration between Image and LiDAR Point Cloud for Vehicle Localization
- LiDAR Registration with Visual Foundation Models
- REGRACE: A Robust and Efficient Graph-based Re-localization Algorithm using Consistency Evaluation
- Adaptive World Memory 3D Foundation Model for Scalable 3D Mapping, Localization, and Rendering
- LetsMap: Unsupervised Representation Learning for Label-Efficient Semantic BEV Mapping
- MapDreamer: Aerial Imagery Conditioned Latent Diffusion for Lane-Level Map Generation
- MRLGraph: A Novel, Graph-Based Map Relation Learning Network for Traffic Element to Lane Assignment Based on Geometric Map Data
- A Good Foundation is Worth Many Labels: Label-Efficient Panoptic Segmentation
- Label-Efficient LiDAR Panoptic Segmentation
- Label-Efficient Point Cloud Segmentation with Active Learning
- BYE: Build Your Encoder with One Sequence of Exploration Data for Long-Term Dynamic Scene Understanding
- Articulated Object Estimation in the Wild
- OGScene3D: Incremental Open-Vocabulary 3D Gaussian Scene Graph Mapping for Scene Understanding
Wolfram Burgard · Learning physical interaction with world models, touch and force
Begin with visual robustness and world-model-based policy learning, then compare tactile prediction, simulation and contact-aware cooperation. Velocity and wrench control connect learned representations to interaction dynamics; the final human-gait study illustrates a related use of motion measurements in biomechanics.
- Augmented Reality for RObots (ARRO): Pointing Visuomotor Policies Towards Visual Robustness
- LUMOS: Language-Conditioned Imitation Learning with World Models
- DiWA: Diffusion Policy Adaptation with World Models
- Goal Achievement Guided Exploitation: A Principled Performance-Based Scheduling Framework for Reinforcement Learning
- Refined Policy Distillation: From VLA Generalists to RL Experts
- Rewarding DINO: Predicting Dense Rewards with Vision Foundation Models
- DUET-DINO: Simultaneous Cross-View World Modeling for Latent Planning in Robot Manipulation
- FlowTouch: View-Invariant Visuo-Tactile Prediction
- Norm2Tex: Augmenting Visuo-Tactile Simulations with Texture
- TACTIC: Understanding Tactile Encoders and Conditioning for Contact-rich Robot Manipulation Policies
- ConPoSe: LLM-Guided Contact Point Selection for Scalable Cooperative Object Pushing
- Shape Formation for the Cooperative Transportation of Arbitrary Objects Using Multi-Agent Reinforcement Learning
- Enabling Dynamic Tracking in Vision-Language-Action Models via Time-Discrete and Time-Continuous Velocity Feedforward
- Wrench-ACT: Enhancing Robot Policies for Contact Rich Behavior Using Direct Wrench Control
- Altered gait stability and regularity in lower limb amputees observed across different body segments
Russ Tedrake · Geometry, search and constrained motion planning
Follow the full constrained-planning pipeline: configuration-space geometry and convex regions, repeated and implicit graph search, hybrid mode sequences, and continuous optimization. Time scaling, inverse kinematics, manifold charts and symmetries expose the assumptions behind efficient geometric planning.
- Non-Euclidean Motion Planning with Graphs of Geodesically-Convex Sets
- Planning Shorter Paths in Graphs of Convex Sets by Undistorting Parametrized Configuration Spaces
- Polytope Inner Approximations with Fewer Faces Subject to Minimum-Volume, Point-Inclusion, and Set-Intersection Constraints
- Faster Algorithms for Growing Collision-Free Convex Polytopes in Robot Configuration Space
- Superfast Configuration-Space Convex Set Computation on GPUs for Online Motion Planning
- Semidefinite Relaxations for Collision-Free Motion Planning
- Multi-Query Shortest-Path Problem in Graphs of Convex Sets
- GCS*: Forward Heuristic Search on Implicit Graphs of Convex Sets
- Mixed Discrete and Continuous Planning using Shortest Walks in Graphs of Convex Sets
- A New Semidefinite Relaxation for Linear and Piecewise-Affine Optimal Control with Time Scaling
- A Framework for Combining Optimization-Based and Analytic Inverse Kinematics
- Planning along Differentiable Charts of Constraint Manifolds with General-Purpose IK Solvers
- Sampling-Based Motion Planning with Discrete Configuration-Space Symmetries
Russ Tedrake · Learning robot control from demonstrations, contact and physics
Start with Diffusion Policy and controlled policy evaluation, then compare data quality, action diversity, constraint fidelity, history and reactive execution. Contact sensing and planning, physics-driven demonstrations and simulation assets lead into the question of which latent state information a controller actually needs.
- Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
- A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation
- Empirical Analysis of Sim-and-Real Cotraining of Diffusion Policies for Planar Pushing from Pixels
- Ambient Diffusion Policy: Imitation Learning from Suboptimal Data in Robotics
- Explicitly capturing demonstrator multi-modality for distillation into diffusion policies
- How Well do Diffusion Policies Learn Kinematic Constraint Manifolds?
- Training and Evaluating Diffusion Policies with Long Context Lengths
- Revisiting Open-Loop Execution in Robotics: Toward Reactive, Higher-Performing Policies
- Proximity and Visuotactile Point Cloud Fusion for Contact Patches in Extreme Deformation
- Dexterous Contact-Rich Manipulation via the Contact Trust Region
- Physics-Driven Data Generation for Contact-Rich Manipulation via Trajectory Optimization
- Scalable Real2Sim: Physics-Aware Asset Generation Via Robotic Pick-and-Place Setups
- Cost-Driven Representation Learning for Linear Quadratic Gaussian Control: Part II
Russ Tedrake · Video world models and simulation for robot learning
Frame the role of data with the robotics debate, then study visual pretraining, history-conditioned video prediction and video-based planning. Controllable scene generation and agent-built simulation environments complete the path from learned worlds to training resources.
- "Data will solve robotics and automation: True or false?": A debate
- Should VLMs be Pre-trained with Image Data?
- History-Guided Video Diffusion
- Large Video Planner Enables Generalizable Robot Control
- Steerable Scene Generation with Post Training and Inference-Time Search
- SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes
Fei-Fei Li · Spatial intelligence, visual memory and physical reasoning
Connect spatial reasoning and cognitive maps with long-video perception, temporal search and fine-grained visual evidence. Active exploration, viewpoint planning and embodied physics benchmarks then test whether visual understanding supports useful action, with MIRAGE providing a critical check on apparent understanding.
- Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces
- MindCube: Spatial Mental Modeling from Limited Views
- Theory of Space: Can Foundation Models Construct Spatial Beliefs through Active Exploration?
- Planning with the Views
- Cambrian-S: Towards Spatial Supersensing in Video
- T*: Re-thinking Temporal Search for Long-Form Video Understanding
- Towards Fine-Grained Video Question Answering
- VideoWeave: A Data-Centric Approach for Efficient Video Understanding
- ESI-Bench: Towards Embodied Spatial Intelligence that Closes the Perception-Action Loop
- BEHAVIOR-Spatial Benchmark: Evaluating Spatial Understanding for Robotics
- ENACT: Evaluating Embodied Cognition with World Modeling of Egocentric Interaction
- QuantiPhy: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models
- MIRAGE: The Illusion of Visual Understanding
Fei-Fei Li · Embodied agents: reasoning, programs and adaptation
Start with interactive agent training and test-time reasoning, then examine multi-turn reinforcement learning, collapse and reflective planning. Coding agents and broad embodied evaluations lead into robot-policy context, compact memory and activation steering as mechanisms for adaptation.
- An Interactive Agent Foundation Model
- s1: Simple test-time scaling
- RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning
- RAGEN-2: Reasoning Collapse in Agentic RL
- VAGEN: Reinforcing World Model Reasoning for Multi-Turn VLM Agents
- Learning from Trials and Errors: Reflective Test-Time Planning for Embodied LLMs
- CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation
- Embodied First Exam
- RoboTTT: Context Scaling for Robot Policies
- T²Mem: Learning Test-Time Memory for Robotics
- SteerAct: Unlocking Action Models through Few-Shot Temporal Steering
Fei-Fei Li · Manipulation from human guidance, rewards and affordances
Follow human demonstrations into whole-body manipulation, generated data, corrections and composable skills. Compare language-derived rewards, preferences, affordances and stereo perception before considering assistive intent and the activities people actually want automated.
- ARCap: Collecting High-quality Human Demonstrations for Robot Learning with Augmented Reality Feedback
- BEHAVIOR Robot Suite: Streamlining Real-World Whole-Body Manipulation for Everyday Household Activities
- MoMaGen: Generating Demonstrations under Soft and Hard Constraints for Multi-Step Bimanual Mobile Manipulation
- I3L: Iterative Interactive Imitation Learning for Whole-Body Mobile Manipulation
- Learning Composable Skills by Discovering Spatial and Temporal Structure with Foundation Models
- GeT-USE: Learning Generalized Tool Usage for Bimanual Mobile Manipulation via Simulated Embodiment Extensions
- A Real-to-Sim-to-Real Approach to Robotic Manipulation with VLM-Generated Iterative Keypoint Rewards
- ROSETTA: Constructing Code-Based Reward from Unconstrained Language Preference
- UAD: Unsupervised Affordance Distillation for Generalization in Robotic Manipulation
- StereoPolicy: Improving Robotic Manipulation Policies via Stereo Perception
- EEG-Based Brain-Computer Interface for Robotic Assistance with User Intention Prediction
- Why Automate This? Exploring the Connection between Time Use, Well-being and Robot Automation Across Social Groups
Fei-Fei Li · Human motion, multimodal demonstrations and dexterous transfer
Begin with representing and reconstructing human motion, including occlusion, then collect creative tool demonstrations. Sound, muscle activity, contact geometry and touch add physical information before force-aware transfer and agent-managed human-to-robot training.
- The Language of Motion: Unifying Verbal and Non-verbal Language of 3D Human Motion
- NeuHMR: Neural Rendering-Guided Human Motion Reconstruction
- Rendering Humans Behind Occlusions
- ToolDex: An Ego-Exo Human Demonstration Dataset for Creative Physical Tool Use
- Chain-of-Modality: Learning Manipulation Programs from Multimodal Human Videos with Vision-Language-Models
- One Demonstration, Many Objects: Generalizing Manipulation via Local Contact Geometry
- T-Rex: Tactile-Reactive Dexterous Manipulation
- FACT: A Multimodal Force-Action-Track Foundation Model for Human-to-Robot Dexterity
- DexAgent: An Agentic Human2Sim2Robot Framework for Dexterous Manipulation with Self-Evolving Tool Library
Fei-Fei Li · Robot world models, physical dynamics and simulation
Compare geometric, video and latent-action interfaces between world prediction and control, including learned object dynamics and test-time adaptation. Then generate assets, affordance annotations and editable simulation scenes, ending with action-preserving rendering for transfer to physical robots.
- Dream2Flow: Bridging Video Generation and Open-World Manipulation with 3D Object Flow
- PointWorld: Scaling 3D World Models for In-The-Wild Robotic Manipulation
- IMPASTO: Integrating Model-Based Planning with Learned Dynamics Models for Robotic Oil Painting Reproduction
- Rapid Adaptation of Particle Dynamics for Generalized Deformable Object Mobile Manipulation
- Masked Visual Actions for Unified World Modeling
- Cross-Embodiment Robot Foundation World Models with Latent Actions
- OpenWAM: An Open Framework for Composable World-Action Models
- TrAct: Bridging Robot Control and Visual Prediction with Visual Tracks
- Dream What I Can: Sparse Visual Planning for Long-Horizon Robot Manipulation
- T2-z: Test-Time Optimization of Environment Latents for Robotic World Models
- Repurposing 2D Diffusion Models with Gaussian Atlas for 3D Generation
- SAGrid: Scaling Robot Simulation through Automatic Affordance Annotation on In-the-Wild 3D Assets
- SimFoundry: Modular and Automated Scene Generation for Policy Learning and Evaluation
- RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer
Fei-Fei Li · Image and video generation: representations, training and evaluation
Trace tokenization, latent computation, language-informed objectives and training data through image generation. Continue into video diffusion, long-horizon consistency, world and motion benchmarks, and preference scoring to distinguish visual quality from useful generative behavior.
- Flow to the Mode: Mode-Seeking Diffusion Autoencoders for State-of-the-Art Image Tokenization
- Latent Forcing: Reordering the Diffusion Trajectory for Pixel-Space Image Generation
- Generative Representation Learning for Scalable Latent Tokenization
- A Language-Informed Loss for Visual Generation
- GPIC: A Giant Permissive Image Corpus for Visual Generation
- Text-to-Image Generative Modeling Without Language Pretraining
- Exploring Diffusion Transformer Designs via Grafting
- Discovering Latent Graphs with GFlowNets for Diverse Conditional Image Generation
- VLIC: Vision-Language Models As Perceptual Judges for Human-Aligned Image Compression
- Photorealistic Video Generation with Diffusion Models
- LoGo: Local-Global Rewards for Consistent Long-Horizon Video Generation
- WorldScore: A Unified Evaluation Benchmark for World Generation
- HumanScore: Benchmarking Human Motions in Generated Videos
- Recreating Video Arenas via Automated Preference Scoring
Fei-Fei Li · Healthcare: ambient sensing and clinical multimodal models
Begin with longitudinal digital biomarkers and observation of hospital care, then move into cardiac-ultrasound segmentation, report assessment and increasingly multimodal interpretation. Compare the clinical evidence and task definitions behind each proposed use.
- A framework of digital biomarkers for neurodegenerative diseases
- Developing ICU Clinical Behavioral Atlas Using Ambient Intelligence and Computer Vision
- TRACER: A Multimodal Harness for Patient Care in Hospitals
- Foundation versus domain-specific models for left ventricular segmentation on cardiac ultrasound
- Evaluating large language models in echocardiography reporting: opportunities and challenges
- EchoGraph system for automated quality assessment of echocardiography reports
- EchoAtlas: A Conversational, Multi-View Vision-Language Foundation Model for Echocardiography Interpretation and Clinical Reasoning
- MARCUS: An agentic, multimodal vision-language model for cardiac diagnosis and management
Fei-Fei Li · AI policy and scientific evidence
Read the analysis of how evidence enters policymaking before the California frontier-AI report. This focused pair addresses scientific uncertainty, institutions and evaluation mechanisms; it remains distinct from healthcare and technical model-building.
Mathematical tools for language models — the complete reading path
A complete sequence from vector geometry and sparse recovery to attention, signal mixing, positional encoding and long context. Continue through optimization geometry, matrix spectra, scaling and evaluation, then policy learning and generative flows. Read the short intuition paragraphs before the formal derivations; each topic explains a different engineering problem rather than one theorem for all of language modeling.
- Extensions of Lipschitz Mappings into a Hilbert Space
- Compressed Sensing
- Robust Uncertainty Principles: Exact Signal Reconstruction from Highly Incomplete Frequency Information
- Toy Models of Superposition
- Towards Monosemanticity: Decomposing Language Models With Dictionary Learning
- Attention Is All You Need
- Hopfield Networks is All You Need
- Concerning nonnegative matrices and doubly stochastic matrices
- mHC: Manifold-Constrained Hyper-Connections
- RoFormer: Enhanced Transformer with Rotary Position Embedding
- Extending Context Window of Large Language Models via Position Interpolation
- YaRN: Efficient Context Window Extension of Large Language Models
- Natural Gradient Works Efficiently in Learning
- Decoupled Weight Decay Regularization
- Shampoo: Preconditioned Stochastic Tensor Optimization
- Muon is Scalable for LLM Training
- Traditional and Heavy-Tailed Self Regularization in Neural Network Models
- Predicting trends in the quality of state-of-the-art neural networks without access to training or testing data
- Scaling Laws for Neural Language Models
- Training Compute-Optimal Large Language Models
- Measuring Massive Multitask Language Understanding
- Training Verifiers to Solve Math Word Problems
- Proximal Policy Optimization Algorithms
- Direct Preference Optimization: Your Language Model is Secretly a Reward Model
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Flow Matching for Generative Modeling
JEPA — image and video representation learning
Follow the visual research line from the original proposal and masked-image prediction through motion, conditioned image transformations, video understanding and intuitive physics. Compare the later dense-feature and efficient-pretraining recipes after their predecessors. The theory and planning sequences develop the mathematical and action-conditioned branches.
- A Path Towards Autonomous Machine Intelligence
- Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
- MC-JEPA: A Joint-Embedding Predictive Architecture for Self-Supervised Learning of Motion and Content Features
- Learning and Leveraging World Models in Visual Representation Learning
- Revisiting Feature Prediction for Learning Visual Representations from Video
- Video Representation Learning with Joint-Embedding Predictive Architectures
- Intuitive physics understanding emerges from self-supervised pretraining on natural videos
- V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
- Back to the Features: DINO as a Foundation for Video World Models
- V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning
- LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
- LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics
JEPA — representation geometry, regularization and identifiability
Start with the energy-based tutorial and the slow-feature counterexample. Then examine density geometry, collapse prevention, sparse representations and latent dynamics, ending with the assumptions needed for identifiability and cross-model alignment. LpWM connects the regularization ideas to planning.
- Introduction to Latent Variable Energy-Based Models: A Path Towards Autonomous Machine Intelligence
- Joint Embedding Predictive Architectures Focus on Slow Features
- Gaussian Embeddings: How JEPAs Secretly Learn Your Data Density
- LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
- Rectified LpJEPA: Joint-Embedding Predictive Architectures with Sparse and Maximum-Entropy Representations
- LpWM: A Case for Sparse Representations in World Models
- When Does LeJEPA Learn a World Model?
- Proving and Improving the Platonic Representation Hypothesis with JEPAs
JEPA — world models for planning and control
Start with latent dynamics learned from offline observations and compare frozen-feature and end-to-end models. Continue through state design, sparse codes and straightened trajectories, then hierarchy, action optimization and online adaptation. Finish with stability limits, structured predictors, quadrotor control and driving. Read results together with their action inputs, planners and evaluation protocols.
- DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models
- V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
- A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures
- LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels
- What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?
- Causal-JEPA: Learning World Models through Object-Level Latent Masking
- LpWM: A Case for Sparse Representations in World Models
- Temporal Straightening for Latent Planning
- Hierarchical Planning with Latent World Models
- H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning
- Value-guided action planning with JEPA world models
- Parallel Stochastic Gradient-Based Planning for World Models
- AdaJEPA: An Adaptive Latent World Model
- Preserving Unstable Modes Through Inverse Dynamics in JEPA World Models
- LeHami: Symplectic Latent Dynamics for World Models
- SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotors
- AD-E2E-JEPA: A Joint-Embedding Predictive Architecture For End-to-End Autonomous Driving
JEPA — language, audio and scientific representations
Compare how views, prediction targets and latent spaces must change when JEPA leaves natural video. Move from paired language/code and semantic trajectories to vision-language answers, speech tokens and soft anchors, then action-conditioned sound, graph partitions, crystals and medical scans. The common questions are what is predicted, what prevents collapse, and where paired data or supervision enters.
- LLM-JEPA: Large Language Models Meet Joint Embedding Predictive Architectures
- Semantic Tube Prediction: Beating LLM Data Efficiency with JEPA
- VL-JEPA: Joint Embedding Predictive Architecture for Vision-language
- JEPA as a Neural Tokenizer: Learning Robust Speech Representations with Density Adaptive Attention
- Soft Clustering Anchors for Self-Supervised Speech Representation Learning in Joint Embedding Prediction Architectures
- S-JEPA : Soft Clustering Anchors for Self-Supervised Speech Representation Learning
- Music-JEPA: Learning a World Model of Sound from Action
- HP-JEPA: Hierarchical Partitioning for Multi-Resolution Graph Joint-Embedding Predictive Learning
- Crys-JEPA: Accelerating Crystal Discovery via Embedding Screening and Generative Refinement
- Self-supervised DXA representations encode multi-system disease risk, biological aging and heritability