Reproduce “R2HandoverSim: A Simulation Framework and Benchmark for Robot-to-Human Object Handovers” 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
Hanxin Zhang, Abdulqader Dhafer, Hongbiao Dong, Zhou Daniel Hao
R2HandoverSim introduces a standardized simulation and evaluation framework for robot-to-human object handovers, combining a fixed UR5e platform, shared object and hand models, motion planning, physics-based execution, and five decomposed quality criteria. Across four baselines, Intent-Handover achieved the highest average simulated success rate (67.9%), the strongest performance on functionally constrained objects (66.0%), and the best physical success rate (73.3%) in a 30-participant, 600-trial study, while user ratings showed that geometric success alone does not reliably indicate comfort or naturalness.
Key insights
- Different handover strategies optimize different aspects of the task: contact-based methods provide strong geometric safety, whereas intent-conditioned generation better captures functional object use and receiver affordance.
- Overall success rate obscures important distinctions; a collision-free handover may still be uncomfortable or obstruct the region the person needs to grasp.
- The static simulated receiver does not fully capture human adaptation, which can improve real-world performance for some methods but also exposes sensitivity to sensor noise, calibration, object weight, and physical execution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.