Literature
Display
← All papers

Universal Agent for Disentangling Environments and Tasks

Jiayuan Mao · Honghua Dong · Joseph J. Lim

· ICLR 2018 ↗

Reading note

Universal Agent divides control into an environment-dependent module for reaching a target state and a task-dependent module for choosing the next target. The simulated experiments test whether this division allows knowledge about movement to be reused when the task changes.

Concept demonstration

Open image ↗