Reproduce “AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics” 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
Xinxin Liu, Bingxin Xu, Wenrui Bao, Ming Li, Wentao Bao, Chen Chen, Yuzhang Shang
AgenticRoboRL trains a language-model planner to use an existing robot harness, keeping its execution components fixed. Supervised fine-tuning on teacher trajectories is followed by multi-turn reinforcement learning through harness rollouts. On LIBERO-PRO, the 9B planner’s success rate on the same RPent harness and frozen VLA rises from 18.1% to 71.0% across 800 episodes, without external memory at rollout or evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.