Reproduce “CANDO: Conservative Action-Noise Directional Optimization for Residual Reinforcement Learning” 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
Zeyu Gao, Chengkai Hou, Chaoyue Li, Zhuangzhe Wu, Jiaming Liu, Yao Mu, Peng Jia, Li Zhang and 1 more
The authors introduce CANDO, which improves a frozen generative robot policy through edits to its input noise and a correction to its generated actions. Its CNDO noise learner uses a conservative critic and a gate to select edits. Across five simulated manipulation tasks and three seeds, CANDO averaged 76.1 NAUC, versus CNDO’s 69.5, but did not lead every task metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.