2023-12-02-jian23a.md

title	section	openreview	abstract	layout	series	publisher	issn	id	month	tex_title	firstpage	lastpage	page	order	cycles	bibtex_author	author	date	address	container-title	volume	genre	issued	pdf	extras
Policy Stitching: Learning Transferable Robot Policies	Poster	2qKBwyLnln	Training robots with reinforcement learning (RL) typically involves heavy interactions with the environment, and the acquired skills are often sensitive to changes in task environments and robot kinematics. Transfer RL aims to leverage previous knowledge to accelerate learning of new tasks or new body configurations. However, existing methods struggle to generalize to novel robot-task combinations and scale to realistic tasks due to complex architecture design or strong regularization that limits the capacity of the learned policy. We propose Policy Stitching, a novel framework that facilitates robot transfer learning for novel combinations of robots and tasks. Our key idea is to apply modular policy design and align the latent representations between the modular interfaces. Our method allows direct stitching of the robot and task modules trained separately to form a new policy for fast adaptation. Our simulated and real-world experiments on various 3D manipulation tasks demonstrate the superior zero-shot and few-shot transfer learning performances of our method.	inproceedings	Proceedings of Machine Learning Research	PMLR	2640-3498	jian23a	0	Policy Stitching: Learning Transferable Robot Policies	3789	3808	3789-3808	3789	false	Jian, Pingcheng and Lee, Easop and Bell, Zachary and Zavlanos, Michael M. and Chen, Boyuan	given family Pingcheng Jian given family Easop Lee given family Zachary Bell given family Michael M. Zavlanos given family Boyuan Chen	2023-12-02		Proceedings of The 7th Conference on Robot Learning	229	inproceedings	date-parts 2023 12 2	https://proceedings.mlr.press/v229/jian23a/jian23a.pdf