arxivcs.ROcs.LG2026-07-05
HRIBench: Benchmarking Interaction-Centric Human-Robot Collaboration
Chang Liu, Jiawei Zhang, Tao Zhang, Ye Wang, Hongyu Zhou, Qin Jin
Current vision-language-action (VLA) benchmarks primarily evaluate isolated manipulation skills while leaving human-robot interaction structure largely unmodeled. However, real-world collaboration fundamentally requires coordination under shared agency, including intent understan…