Reproduce “Event2Reward: Learning Robotic Rewards from Progress Relations” 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

Junnan Nie, Zitian Fan, Jiachen Zhang, Junyi Lao, Likui Zhang, Chenghao Liu, Xiaoyang Lu, Yunbo Wang and 9 more

Event2Reward combines normalized time labels from demonstrations with local event-based comparisons from robot rollouts to estimate task progress and shape rewards. In a controlled ablation across four LIBERO-10 tasks, it achieved 36.7% mean normalized evaluation AUC, versus 19.4% with trajectory preferences and 19.1% without relation loss. These results support the relation objective in this tested setting, not universally.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.