Reproduce “To Do or Not to Do: Ensuring the Safety of Visuomotor Policies Learned from Demonstrations” 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
Riad Ahmed, Moniruzzaman Akash, Momotaz Begum
The authors introduce runtime recovery for visual imitation-learning policies, keeping the robot in a task-specific region where the goal is visible and recognizable. It checks visibility and familiarity, then adjusts actions without a dynamics model or retraining. In candy sorting under combined distribution shifts, success rose from 0.42 ± 0.05 to 0.84 ± 0.01; these empirical rates are not a universal guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.