Reproduce “HARP-Bench: Benchmarking Robot Manipulation Around Active 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
Qinhong Zhou, Riley Zong, Chunru Lin, Diyi Hu, Shivraj Singh Bhatti, Hongxin Zhang, Lin Qiu, Hanqing Zeng and 1 more
HARP-Bench evaluates manipulation policies during scripted human-avatar events that alter goals, progress, object availability, or movement constraints. It organizes 37 task-condition instances into four categories and tests policies in closed loop. Among learned policies, π₀ has highest all-task success (0.19), while ACT scores lower (0.10) but better on all three geometric safety measures, which do not establish deployment safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.