Reproduce “Labimus: A Simulation and Benchmark for Humanoid Dexterous Manipulation in Chemical Laboratory” 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
Yuhan Wu, Zhao Jin, Tao Li, Yuheng Zhang, Zhichao Wang, Shuo Wang, Jun Jiang, Xiaobo Li and 4 more
Labimus provides the first simulation and benchmark platform for evaluating humanoid dexterous manipulation in precision-critical chemical laboratory tasks, revealing a "precision gap" where current robotic methods often complete tasks but fail to meet scientific quantitative tolerances. This platform reconstructs over 30 laboratory assets and simulates granular materials to enable realistic assessment of complex operations like solid weighing.
Key insights
- A significant "precision gap" exists where policies achieve binary task completion but fail to meet the continuous quantitative tolerances required for scientific experimental validity.
- Fine, single-finger manipulation tasks, such as pressing a small button, present a major challenge for current imitation learning and vision-language-action policies, resulting in near-zero success rates.
- While policies show relative insensitivity to individual visual perturbations, compounded visual variations (lighting and texture combined) lead to a more substantial performance degradation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.