Literature
Display
← All papers

Robust Robot Walker: Learning Agile Locomotion over Tiny Traps

Shaoting Zhu · Runhan Huang · Linzhan Mou · Hang Zhao

· ICRA 2025 ↗

Robotics9 min text · CondensedOriginal paper ↗
Contents

At a glance

Robust Robot Walker trains a Unitree A1 quadruped to cross small bars, pits, and poles using proprioception at deployment. Simulation supplies privileged contact forces, velocity, friction, and trap labels to train a contact representation and a recurrent estimator. A second stage gradually replaces privileged information with estimates.

The method combines dense goal-tracking rewards with constant, artificial goal commands for teleoperation. It reaches 93.4% success on a mixed simulated trap course and demonstrates physical recovery from several trap types. The controller reacts after contact rather than visually anticipating hazards. Its approximate omnidirectional interface cannot independently command speed magnitude, and highly deformable traps remain difficult.

Task overview

The task is reactive traversal of bars, pits, and poles using the robot’s own state rather than camera input.

Refer to caption
Fig. 1: Our Robust Robot Walker is passing through various challenging “tiny traps” including Pit, Bar, and Pole solely relying on its proprioception.

I INTRODUCTION

Thin or occluded hazards can be unreliable in depth images and may lie below or behind the camera. The proposed response is to infer contact consequences from the robot’s own motion and joint signals. Figure 2 illustrates depth-sensing failures; it does not imply that all external perception is ineffective.

The key components are a learned contact-force latent, trap-class supervision during training, probability-annealed replacement of privileged state, and dense goal-tracking rewards. Artificial goal vectors turn the trained goal policy into a joystick-controlled movement policy without motion capture or external localization.

Refer to caption
Fig. 2: Camera unreliability in fine-grained trap scenarios.
Refer to caption
Fig. 3: The training and deployment overview of Robust Robot Walker. Our method achieves explicit-implicit dual-state estimation and approximate omnidirectional movement. Each color on the quadruped robot corresponds to a type of joint link: Orange-Base, Yellow-Hip, Blue-Thigh, Purple-Calf, and Foot (Unseen).

II RELATED WORKS

The method builds on reinforcement-learning locomotion and learned contact estimation, combining them for small hazards that the robot encounters before it can react.

II-A Learning-Based Locomotion

Prior learned controllers cover uneven terrain, stairs, parkour, and manipulation with legs. Goal-driven locomotion and agile blind locomotion motivate the task formulation here. The contribution targets trap recovery and deployment commands, rather than proposing reinforcement learning for locomotion itself.

II-B Collision Detection

Collision handling includes detection, localization, force identification, classification, and reaction. Model-based observers infer contact from dynamics; model-free methods learn explicit contact states or implicit representations. The comparison with Boolean collision prediction motivates a continuous contact-force latent, though its transfer benefit is assessed within this training system.

III METHOD

Training first exposes the policy to privileged simulation signals, then adapts it to a recurrent state estimate. Deployment retains only proprioception, recurrent memory, and operator commands.

III-A Task Definition

Bars are horizontal obstacles below head height, pits are depressions that catch a foot, and poles are thin upright obstacles. The robot must pass them under constant commands while adapting its own motion. Both thin and thicker obstacles appear in the evaluation; the thicker ones also test behavioral generalization even when a camera could detect them.

Refer to caption
Fig. 4: Three categories of tiny traps: Bar, Pit, and Pole.

III-B Reinforcement Learning Setting

PPO controls twelve desired joint positions every 0.02 seconds. Proprioception contains gravity direction, base angular velocity, joint positions and velocities, and the previous action. Training additionally provides base linear velocity, ground friction, and clipped/normalized contact forces for the robot links.

A training command specifies a goal displacement in the robot frame and remaining episode time. The world-frame goal stays fixed while its relative displacement updates with robot motion; vertical displacement is zero. Within 0.2 m, the displacement is set to zero to train stopping. Deployment replaces this evolving localization-dependent command with constant artificial values.

III-C Reward Function

The task reward increases densely as distance to the goal shrinks. A heading term encourages facing the goal, simplifying trap learning mainly to forward or slightly angled approaches. Near the goal, negative-weight finish terms penalize joint deviation from the default pose and residual linear/angular speed.

Regularization discourages stalling and excessive motion while encouraging stable, smooth gait; an adversarial motion prior adds a style reward. Table III retains every formula and weight. The velocity-limit term is a training incentive, not a hard safety constraint. Its source expression uses signed yaw velocity rather than absolute yaw speed; the formula is preserved as printed.

Equations 1–6
rg​o​a​l=10.4+‖Δ​G‖2,r_{goal}=\frac{1}{0.4+\|\Delta G\|_{2}}, (1)
rh​e​a​d​i​n​g={Δ​x+ϵ‖Δ​G‖2+ϵ,‖Δ​G‖2≠01,‖Δ​G‖2=0,r_{heading}=\left\{\begin{matrix}\vskip 5.69054pt\displaystyle\frac{\Delta x+\epsilon}{\|\Delta G\|_{2}+\epsilon},&\|\Delta G\|_{2}\neq 0\\ 1,&\|\Delta G\|_{2}=0\\ \end{matrix}\right., (2)
rf​i​n​i​s​h​_​p​o​s=∑i=112|q−qd​e​f​a​u​l​t|,r_{finish\_pos}=\sum^{12}_{i=1}{{\lvert q-q_{default}\rvert}},\\ (3)
rf​i​n​i​s​h​_​v​e​l=‖v‖+‖ω‖,r_{finish\_vel}=\|v\|+\|\omega\|, (4)
rf​i​n​i​s​h=λ1⋅rf​i​n​i​s​h​_​p​o​s+λ2⋅rf​i​n​i​s​h​_​v​e​l.r_{finish}=\lambda_{1}\cdot r_{finish\_pos}+\lambda_{2}\cdot r_{finish\_vel}. (5)
rv​e​l​_​l​i​m​i​t=(ωz<ωl​i​m​i​t)⋅(‖vx,y‖2<vl​i​m​i​t)r_{vel\_limit}=(\omega_{z}<\omega_{limit})\cdot(\|v_{x,y}\|_{2}<v_{limit})\vskip-5.69054pt (6)

Refer to caption

Fig. 5: Tiny Trap Benchmark. Robots begin on the left side and must pass through tiny traps to reach the goal on the right side.

III-D Training and Deployment

Stage one encodes simulated contact forces into a latent, concatenates explicit velocity/friction state, and trains the policy with those privileged quantities. A trap-classification head shapes the contact representation. A recurrent estimator simultaneously learns to reconstruct the privileged latent from proprioception and commands.

Stage two initializes the estimator and low-level recurrent policy from stage one, then uses probability annealing to transition from privileged inputs to predicted state. Equations 7–11 retain the losses and selection rule. Training uses 4,096 parallel robots and shortens episodes from twenty to eight seconds. The main text specifies fifty rollout steps, whereas Table VII lists twenty-four; the discrepancy remains unresolved in the source.

At deployment, a joystick supplies a constant goal direction, distance, and remaining-time value. A small displacement just above 0.2 m produces mostly translation; a larger lateral displacement produces turning. The policy can stop and switch commands without resets. This gives approximate directional and turning control, not independent tracking of arbitrary forward, lateral, and angular speeds.

Equations 7–11
Ls​u​r​r​o​g​a​t​e+Lv​a​l​u​e+Lr​e​c​o​n+Lc​l​a​s​s​i​f​y+Ld​i​s​c​r​i​m​i​n​a​t​o​r.L_{surrogate}+L_{value}+L_{recon}+L_{classify}+L_{discriminator}.\vskip-2.84526pt (7)
𝒍t\displaystyle\boldsymbol{l}_{t} =Estimator​(𝒑t,𝒈t),\displaystyle=\text{Estimator}(\boldsymbol{p}_{t},\boldsymbol{g}_{t}), (8)
𝒊t\displaystyle\boldsymbol{i}_{t} =Probability Selection ​(𝑷t,𝒍t,𝒍t^),\displaystyle=\text{Probability Selection~}(\boldsymbol{P}_{t},\boldsymbol{l}_{t},\hat{\boldsymbol{l}_{t}}), (9)
𝒂t\displaystyle\boldsymbol{a}_{t} =Actor​El​o​w​(𝒊t,𝒑t,𝒈t),\displaystyle=\text{Actor}~E_{low}(\boldsymbol{i}_{t},\boldsymbol{p}_{t},\boldsymbol{g}_{t}), (10)
Probability ​𝑷t\displaystyle\text{Probability~}\boldsymbol{P}_{t} =𝜶i​t​e​r​a​t​i​o​n.\displaystyle=\boldsymbol{\alpha}^{iteration}. (11)

IV EXPERIMENTS

Experiments combine long simulated courses, contact-input ablations, twenty-trial physical comparisons per test setting, and simulation analyses of commands and representations.

IV-A Experiment Setup

Isaac Gym trains 4,096 agents on one RTX 3090: 12,000 iterations on plane terrain followed by 30,000 trap-terrain iterations for the two training stages as described. Both simulation and hardware run the policy at 50 Hz. The A1 uses a Jetson Xavier NX, ROS teleoperation messages, and joint PD control with Kp = 40 and Kd = 0.5.

TABLE I: The comparison results with other locomotion policies in our benchmark.
Method Success Rate ↑\uparrow Average Pass Time (s) ↓\downarrow Average Travel Distance (m) ↑\uparrow
Mix Bar Pit Pole Mix Bar Pit Pole Mix Bar Pit Pole
Ours 0.934 0.941 0.902 0.702 119.03 130.65 111.91 156.40 56.37 55.54 56.22 44.06
w/o goal command 0.000 0.000 0.000 0.279 300.00 300.00 300.00 254.17 3.41 3.81 11.32 24.15
Ours w/ Boolean 0.779 0.718 0.656 0.658 144.66 169.59 161.17 167.21 52.40 50.12 49.28 41.65
RMA 0.000 0.000 0.000 0.067 300.00 300.00 300.00 291.99 5.27 7.45 10.21 17.17
MoB 0.000 0.000 0.000 0.000 300.00 300.00 300.00 300.00 2.90 4.13 4.42 6.57
HIMLoco 0.000 0.000 0.000 0.000 300.00 300.00 300.00 300.00 3.61 4.16 16.73 16.15
TABLE II: The ablation results of different link. B-base link, H-hip link, T-thigh link, C-calf link, and F-foot link.
Method Success Rate ↑\uparrow Average Pass Time (s) ↓\downarrow Average Travel Distance (m) ↑\uparrow
Mix Bar Pit Pole Mix Bar Pit Pole Mix Bar Pit Pole
BHTCF (Ours) 0.934 0.941 0.902 0.702 119.03 130.65 111.91 156.40 56.37 55.54 56.22 44.06
BTCF 0.918 0.925 0.901 0.704 122.58 138.98 111.99 158.91 56.21 55.10 55.97 41.59
TCF 0.828 0.815 0.820 0.618 144.47 163.72 134.32 176.16 53.03 48.44 52.62 37.77
TC 0.904 0.929 0.902 0.687 120.88 125.19 113.36 159.21 55.58 54.89 54.96 41.54
BTC 0.901 0.902 0.899 0.729 123.59 136.90 112.15 153.41 55.70 53.18 54.17 46.03
T 0.864 0.919 0.850 0.692 128.97 128.28 122.87 159.60 53.97 53.44 53.23 41.87
Prop (None) 0.819 0.874 0.796 0.655 142.84 148.01 139.03 164.72 52.10 50.02 49.85 39.40
Refer to caption
Fig. 6: Real-world experiments. The height of the bar is between the range [0.1m, 0.25m]. The width of the pit is between the range [0.1m, 0.3m]. The width of the pole is between the range [0.01m, 0.13m].

IV-B Tiny Trap Benchmark

The mixed benchmark is a 5×60 m runway containing ten bars, fifty poles, and ten pits. Bar height and pit width range from 0.05–0.20 m. Separate single-trap courses triple the number of their respective obstacles. Each test uses 1,000 robots and the deployment-style artificial goal commands.

Success means reaching within 0.2 m of the target in 300 seconds. Falling, rolling, or remaining stuck counts as failure. Failed trials contribute the full 300 seconds to mean pass time, so that metric is not successful-traversal time alone. The source also averages traveled distance over all robots.

IV-C Simulation Experiments

The full method scores 93.4%, 94.1%, 90.2%, and 70.2% success on mixed, bar, pit, and pole courses. Boolean contact inputs reduce those values to 77.9%, 71.8%, 65.6%, and 65.8%. The velocity-command variant fails the first three long courses but reaches 27.9% on poles.

RMA, MoB, and HIMLoco perform poorly on this benchmark. RMA and HIMLoco are retrained on trap terrain while retaining their original rewards and random-velocity command settings, which the authors acknowledge may disadvantage them. These results compare the complete configured systems rather than isolating contact representation alone.

Using all links gives the best mixed-course score, compared with 81.9% for the proprioception-only training ablation. The ablation is not monotonic across every task: base/thigh/calf inputs reach 72.9% on poles, exceeding the full set’s 70.2%, and thigh/calf inputs tie the full method on pits. Table II supports useful contact information without proving each added link always improves performance.

IV-D Real-world Experiments

Physical behavior includes backing and lifting legs over bars, redistributing support or kicking free from pits, and sidestepping poles. Each plotted setting uses twenty repeated trials. Figure 6 compares the proposed method, a forward-velocity variant, and the default MPC controller; the proposed method performs best overall, while the velocity variant is much closer than on the long simulated courses.

Performance declines on the largest tested bar and pit settings. Figure 7 retains successful sequences, and Figure 9 retains a failure of the Boolean-contact policy after stopping and reversing on flat ground. The authors interpret that failure as a simulation-to-real mismatch in a sparse contact representation; the continuous-force encoder reduces the observed problem, without establishing that transfer errors disappear.

Refer to caption
Fig. 7: The quadruped robot is passing through tiny traps.
Refer to caption
Fig. 8: Experiment results of different Δ​G\Delta G. Different goal distances and directions can cause different movements.
Refer to caption
Fig. 9: The robot falls down on the ground.

IV-E Additional Experiments

Simulation sweeps show that a constant remaining-time command of about three to five seconds works best for crossing one 0.2 m bar. Extreme values degrade performance. Sweeping goal distance and direction changes translation and turning, but the resulting speed magnitude remains constrained by the reward design.

t-SNE plots make the continuous-force/classification representation look more separated and continuous than the Boolean variants. They are qualitative visualizations, not an independent test of contact-estimation accuracy. A normalized Jacobian analysis assigns roughly one quarter of grouped input sensitivity to contact information and finds the base link largest within the contact group. Sensitivity depends on input scaling and is not a causal ablation.

Refer to caption
Fig. 10: Experiment results of different Δ​t\Delta t. The fake goal command is comparable to or better than the real one.
Refer to caption
Fig. 11: T-SNE results. Middle: using the boolean collision state and an encoder without classification loss. Right: only using boolean collision without encoder and classification loss.
Refer to caption
Fig. 12: Importance analysis on latent space and contact links.

V CONCLUSION

The paper demonstrates contact-informed recovery over several small traps with a proprioceptive policy and a simple teleoperation interface. Practical limits include reacting only after contact, limited control of speed magnitude, training’s emphasis on forward-facing crossings, and difficulty with highly deformable obstacles such as rubber bands. Velocity-control changes and real-world finetuning are proposed future work.

APPENDIX

Appendices A–G provide reward formulas, recurrent network sizes, randomized dynamics and observation noise, terrain curriculum, optimization settings, sensitivity equations, and physical/low-light demonstrations.

TABLE III: Reward functions
Type Item Formula Weight
Task Get goal 10.4+‖Δ​G‖2\displaystyle\frac{1}{0.4+\|\Delta G\|_{2}} 5.0
Heading {Δ​x‖Δ​G‖2+ϵ,‖Δ​G‖2≠01,‖Δ​G‖2=0\displaystyle\left\{\begin{matrix}\vskip 5.69054pt\displaystyle\frac{\Delta x}{\|\Delta G\|_{2}+\epsilon},&\|\Delta G\|_{2}\neq 0\\ 1,&\|\Delta G\|_{2}=0\\ \end{matrix}\right. 3.03.0
Finish vel (‖v‖+‖ω‖)⋅(‖Δ​G‖2<0.2)(\displaystyle\|v\|+\|\omega\|)\cdot(\|\Delta G\|_{2}<0.2) -1.0
Finish pos (∑i=112|q−qd​e​f​a​u​l​t|)⋅(‖Δ​G‖2<0.2)\displaystyle(\sum^{12}_{i=1}{\lvert q-q_{default}\rvert)\cdot(\|\Delta G\|_{2}<0.2)} -1.0
Alive 1 3.03.0
Regularization Stall (‖Vx,y‖2<0.1)⋅(‖Δ​G‖2>0.25)(\|V_{x,y}\|_{2}<0.1)\cdot(\|\Delta G\|_{2}>0.25) −2.0-2.0
Vel limit (ωz<ωl​i​m​i​t)⋅(‖vx,y‖2<vl​i​m​i​t)(\omega_{z}<\omega_{limit})\cdot(\|v_{x,y}\|_{2}<v_{limit}) 2.02.0
Joint vel −‖𝐪˙‖2-\|\dot{\mathbf{q}}\|_{2} 0.0020.002
Joint acc −‖𝐪¨‖2-\|\ddot{\mathbf{q}}\|_{2} 2×10−62\times 10^{-6}
Ang vel stability −(‖ωt,x‖2+‖ωt,y‖2)-(\|\omega_{t,x}\|_{2}+\|\omega_{t,y}\|_{2}) 0.20.2
Feet in air ∑i=03(𝐭a​i​r,i−0.3)+10⋅max⁡(0.5−𝐭a​i​r,i,0)\displaystyle\sum_{i=0}^{3}\left(\mathbf{t}_{air,i}-0.3\right)+10\cdot\max\left(0.5-\mathbf{t}_{air,i},0\right) 0.050.05
Balance ‖Ff​e​e​t,0+Ff​e​e​t,2−Ff​e​e​t,1−Ff​e​e​t,3‖2\|F_{feet,0}+F_{feet,2}-F_{feet,1}-F_{feet,3}\|_{2} −2×10−5-2\times 10^{-5}
Style AMP max⁡[0,1−0.25​(𝒟a​m​p​(st,st+1)−1)2]\max\left[0,1-0.25\left(\mathcal{D}_{amp}(s_{t},s_{t+1})-1\right)^{2}\right] 0.1

V-A Reward Functions

Table III specifies the full weighted reward. Stall and finish terms discourage unwanted immobility or motion; joint and angular-velocity terms regularize smoothness; air-time and balance terms shape gait. The heading formula differs slightly from Equation 2 by omitting the numerator’s epsilon, and both printed versions are retained.

The style reference contains 19-dimensional state transitions generated by an MPC controller: joint positions, base height, linear velocity, and angular velocity. Two hundred velocity commands run for two seconds each. A least-squares discriminator with a gradient penalty distinguishes reference from policy transitions, and its score yields Equation 12’s style reward.

Equations 12
Ld​i​s​c​r​i​m​i​n​a​t​o​r\displaystyle L_{discriminator} =𝔼(st,st+1)∼MPC​[(𝒟a​m​p​(st,st+1)−1)2]\displaystyle=\mathbb{E}_{(s_{t},s_{t+1})\sim\text{MPC}}\left[\left(\mathcal{D}_{amp}\left(s_{t},s_{t+1}\right)-1\right)^{2}\right]
+𝔼(st,st+1)∼Policy​[(𝒟a​m​p​(st,st+1)+1)2]\displaystyle+\mathbb{E}_{(s_{t},s_{t+1})\sim\text{Policy}}\left[\left(\mathcal{D}_{amp}\left(s_{t},s_{t+1}\right)+1\right)^{2}\right]
+αg​p​𝔼(st,st+1)∼MPC​[‖∇𝒟a​m​p​(st,st+1)‖2],\displaystyle+\alpha^{gp}\mathbb{E}_{(s_{t},s_{t+1})\sim\text{MPC}}\begin{bmatrix}\left\|\nabla\mathcal{D}_{amp}\left(s_{t},s_{t+1}\right)\right\|_{2}\end{bmatrix},
rs​t​y​l​e=max⁡[0,1−0.25​(𝒟a​m​p​(st,st+1)−1)2].r_{style}=\max\left[0,1-0.25\left(\mathcal{D}_{amp}(s_{t},s_{t+1})-1\right)^{2}\right]. (12)

V-B Network Architectures

Actor and critic are LSTMs with hidden layers [512, 256]. The estimator is an LSTM with [256, 256], followed by a [256, 256] latent MLP. The contact encoder uses [32, 16]. Table IV retains each network’s input and output; privileged critic/contact inputs are training signals rather than deployed sensors.

TABLE IV: Network architecture details
Network Type Input Hidden layers Output
Actor RNN LSTM 𝒑t\boldsymbol{p}_{t}, 𝒍t\boldsymbol{l}_{t}, 𝒈t\boldsymbol{g}_{t} [512, 256] 𝒂t\boldsymbol{a}_{t}
Critic RNN LSTM 𝒑t\boldsymbol{p}_{t}, 𝒔^t\hat{\boldsymbol{s}}_{t}, 𝒄t^\hat{\boldsymbol{c}_{t}}, 𝒈t\boldsymbol{g}_{t} [512, 256] 𝑽t\boldsymbol{V}_{t}
Estimator Module LSTM 𝒑t\boldsymbol{p}_{t}, 𝒈t\boldsymbol{g}_{t} [256, 256] 𝒉t\boldsymbol{h}_{t}
Latent Encoder MLP 𝒉t\boldsymbol{h}_{t} [256, 256] 𝒍t\boldsymbol{l}_{t}
Contact Encoder MLP 𝒄t^\hat{\boldsymbol{c}_{t}} [32, 16] 𝒍tc\boldsymbol{l}_{t_{c}}

V-C Dynamic Randomization

Randomization covers base mass, center-of-mass offsets, friction, initial joints and velocity, motor strength, and proprioception latency. Table V prints a base-mass range of 0–3 kg without clearly identifying whether it is an offset, and a latency range of 0.2–0.4 seconds; these values are retained without reinterpretation.

Table VI adds observation noise to velocity, gravity, and joint signals. The paper attributes improved transfer to these variations, but does not isolate each randomization source’s effect.

TABLE V: Domain randomization
Parameters Range Unit
Base mass [0, 3] k​gkg
Mass position of X axis [-0.2, 0.2] mm
Mass position of Y axis [-0.1, 0.1] mm
Mass position of Z axis [-0.05, 0.05] mm
Friction [0, 2] -
Initial joint positions [0.5, 1.5]×~\times~nominal value r​a​drad
Initial base velocity [-1.0, 1.0] (all directions) m/sm/s
Motor strength [0.9, 1.1]×~\times~nominal value -
Proprioception latency [0.2, 0.4] ss
TABLE VI: Gaussian noise
Observation Gaussian Noise Amplitude Unit
Linear velocity 0.05 m/sm/s
Angular velocity 0.2 r​a​d/srad/s
Gravity 0.05 m/s2m/s^{2}
Joint position 0.01 r​a​drad
Joint velocity 1.5 r​a​d/srad/s

V-D Trap Terrain Setting in Simulation

A 10×10 terrain grid allocates three columns to bars, two to pits, three to poles, and two to planes, with ten difficulty levels. Circular bars and pits prevent bypassing the obstacle. Bar heights grow from 0.05–0.25 m, pit widths from 0.05–0.30 m, and pole counts from ten to sixty; bar/pole widths vary from 0.025–0.10 m. Perlin-noise amplitudes range from 0.05–0.15 m.

Refer to caption
Fig. 13: Trap Terrain setting.

V-E Training Hyperparameters

Table VII retains PPO settings, including learning rate 0.001, clipping 0.2, discount 0.99, GAE parameter 0.95, and four minibatches. Its twenty-four rollout steps differ from the main text’s fifty. Contact forces are clipped to 0–100 N before the reported normalization. Table VIII gives t-SNE initialization, perplexity thirty, and learning rate two hundred.

TABLE VII: PPO hyperparameters
   Hyperparameter       Value   
   clip min std       0.05   
   clip param       0.2   
   desired kl       0.01   
   entropy coef       0.01   
   gamma       0.99   
   lam       0.95   
   learning rate       0.001   
   max grad norm       1   
   num mini batch       4   
   num steps per env       24   
TABLE VIII: T-SNE hyperparameters
   Hyperparameter       Value   
   init       ‘random’   
   perplexity       30   
   learning rate       200   

V-F Importance Analysis

The analysis differentiates actions with respect to inputs, takes absolute derivatives, scales each input by its specified range, sums across actions, then averages and normalizes by input group. Equations 13–18 retain the calculation. The final denominator uses an N-indexed sum although the text defines k comparison groups; the original expression remains unchanged. These normalized derivatives measure local model sensitivity under the chosen scaling.

Equations 13–18
O=Policy​(I),O=\text{Policy}(I), (13)
J=[∂O1∂I1∂O1∂I2⋯∂O1∂In∂O2∂I1∂O2∂I2⋯∂O2∂In⋮⋮⋮⋮∂Om∂I1∂Om∂I2⋯∂Om∂In],J=\begin{bmatrix}\vskip 6.45831pt\dfrac{\partial O_{1}}{\partial I_{1}}&\dfrac{\partial O_{1}}{\partial I_{2}}&\cdots&\dfrac{\partial O_{1}}{\partial I_{n}}\\ \vskip 6.45831pt\dfrac{\partial O_{2}}{\partial I_{1}}&\dfrac{\partial O_{2}}{\partial I_{2}}&\cdots&\dfrac{\partial O_{2}}{\partial I_{n}}\\ \vskip 6.45831pt\vdots&\vdots&\vdots&\vdots\\ \vskip 6.45831pt\dfrac{\partial O_{m}}{\partial I_{1}}&\dfrac{\partial O_{m}}{\partial I_{2}}&\cdots&\dfrac{\partial O_{m}}{\partial I_{n}}\end{bmatrix}, (14)
Ja​b​s=|J|,J_{abs}=\lvert J\rvert, (15)
Si=(Ui−Li)⋅∑j=1MJa​b​s(j,i),i=1,2,⋯,NS_{i}=(U_{i}-L_{i})\cdot\sum_{j=1}^{M}J_{abs}(j,i),~~i=1,2,\cdots,N (16)
S𝒢=∑Ii∈𝒢Sinum​(Ii∈𝒢),S_{\mathcal{G}}=\frac{\sum_{I_{i}\in\mathcal{G}}S_{i}}{\text{num}(I_{i}\in\mathcal{G})}, (17)
Ri=S𝒢i∑j=1NS𝒢j,i=1,2,⋯,kR_{i}=\frac{S_{\mathcal{G}_{i}}}{\sum_{j=1}^{N}S_{\mathcal{G}_{j}}},~~i=1,2,\cdots,k (18)

V-G Real-world Experiment Settings and Additional Results

Physical traps include 6 mm and 20 mm bars, 8 mm table-leg poles, larger poles/sticks, and gaps between 40 mm-high wooden boxes. The photographed objects and all low-light crossing sequences are retained. The demonstration dimensions differ slightly from Figure 6’s rounded plotted pole-width range.

The robot also crosses bars, pits, and poles in low light without visual input. These sequences illustrate the sensor choice’s usefulness in darkness; the appendix does not report a separate low-light trial count or aggregate success rate.

Refer to caption
Fig. 14: Real-world traps: Thick Bar, Thin Bar, Thin Pole, Thick Pole, Pit.
Refer to caption
Fig. 15: Crossing Bar in low-light environment.
Refer to caption
Fig. 16: Crossing Pit in low-light environment.
Refer to caption
Fig. 17: Crossing Pole in low-light environment.
Concept demonstration

Open image ↗