Reproduce “Towards Balanced Behavior Cloning from Imbalanced Datasets” 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
Sagar Parekh, Heramb Nemlekar, Dylan P. Losey
The authors analyze how imbalanced demonstrations affect robot behavior cloning and propose meta-gradient reweighting. The method estimates a best-achievable imitation-loss target for each behavior, then adjusts training weights relative to those targets. In CALVIN picking tasks, underrepresented behaviors perform worse than balanced-data policies; the proposed method improves results in some picking comparisons, but opening-task gains are not statistically significant.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.