Reproduce “Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation” 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
Xiaoman Li, Zhiqiang Yang, Jiayu Hu, Jipeng Kong, Yang Zhang, Jingwei Zhang, Fangzheng Yan, Chunhua Shen and 1 more
GR-BC scores observed action chunks by estimating how execution changes remaining graph distance to success. It builds a graph from demonstrations and rollouts, searches backward from successful endings, and weights chunks by progress while limiting regressions. In a controlled Graphics Card study, X-VLA with GR-BC reached 92% success after round 3, versus 63% for uniform aggregation. Estimates apply within collected-data support.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.