Reproduce “SimPT: An Automated Simulation Framework for Intervention-Based VLA Post-Training” 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
Caicheng Wang, Zipeng Ye, Zhexuan Zhou, Zilong Huang, Xi Zhang, Yuchen Xie
SimPT automates simulated corrective-data collection and evaluation for vision-language-action policies. Its Progress Graph tracks task stages using privileged simulator state, while a motion-planning expert attempts recovery from saved snapshots. Across seven RoboTwin tasks, recovery attempts fell from 3,781 to 945 and recovery success rose from 93.4% to 98.9%. Evaluation runs the updated policy alone; real-robot post-training remained human-in-the-loop.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.