Event

Xuanye Wu et al. Release DexBioLab, a Dexterous Manipulation Simulation Benchmark for Biological Laboratories

1 reports · 1 sources

Xuanye Wu and other researchers proposed and released DexBioLab, a dexterous manipulation simulation benchmark and framework for real biological laboratory environments, containing 14 tasks and 4 robot embodiments, covering functional tool use, fine manipulation, language-conditioned visual grounding, and instrument operation. The framework incorporates controllable dynamic experimental phenomena (flames, steam, bubbles) and adopts a hybrid rendering pipeline: interactive simulation of foreground objects combined with Gaussian-splatting backgrounds reconstructed from 6 real laboratory scenes. The researchers state that real-world experiments show a narrowed sim-to-real gap, supporting the effectiveness of AutoBio+ in real manipulation evaluation.