Reproduce “CARE: Experience-Guided Atomic Corrective Execution for Vision-Language-Action Policies” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junlan Xiao, Junwei Jiang, Zaibin Zhang, Yifan Wang, Zhongbo Zhang, Huchuan Lu, Lijun Wang
CARE trains a vision-language-action policy on corrective examples synthesized from execution failures and uses a 3D monitor to select adjustments or repeat failed atomic actions. It models stage-relative pose deviations and verifies geometric conditions during execution. On a dual-arm SO-101 robot, CARE raised average success across four tasks to 50.0%, versus 30.8% for the base policy, under the tested protocol.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.