Reproduce “Training-free Behavior Cloning” 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
Maximilian Adang, Timothy Chen, Lars Osterberg, Aiden Swann, Mac Schwager
Behavior Predictive Control uses recent robot history to retrieve demonstration windows, reconstruct that history with fitted coefficients, continue their actions, and add a learned correction. On xArm Switch, it succeeded in 34 of 40 trials, versus 8 of 40 for fine-tuned π0.5. Across eight MimicGen tasks, it averaged 73% versus 77%, so results are competitive rather than universally superior.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.