ACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation Models
ACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation Models
This paper presents ACCRUE, a continual off-policy reinforcement learning framework designed to improve robot foundation models by enabling efficient learning of new tasks while retaining previously acquired behaviors. The approach uses lightweight actor and critic networks, distillation of prior behavior, and fixed visual features to maintain compatibility with base-policy updates, achieving near-zero forgetting and reducing real-robot training time by 31.0% to 68.5%.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
% to 68.5\%.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org