Reproduce “BATCH: Behavior Alignment for Cross-Embodiment Learning from Humans” 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
Zeyi Liu, Shuran Song
BATCH matches corresponding stages in human and robot demonstrations using temporal features and subsequence dynamic time warping. It uses aligned pairs for co-training and to condition robot-video prediction on human clips, while deployment uses only robot video. On Cup on Saucer, BATCH achieves 83.3% success across 30 out-of-domain trials, versus 50.0% with random pairing, supporting alignment in these tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.