An Embodied Simulation Platform, Benchmark, and Data-Efficient Augmentation Framework for Wet-Lab Robotics

Pipette, an embodied simulation platform for wet-lab robotics, has been released, featuring over 43 open-source, re-editable assets and an extensible asset-building pipeline. It incorporates a simulation-based data augmentation pipeline that enhances training data from limited demonstrations through various perturbations, achieving a 65.5% average success rate across 11 tasks with only 30 demonstrations each, and improving model performance for SmolVLA and {\pi}0. This framework facilitates natural-language-driven task registration, making it accessible for non-expert users and advancing the efficiency of training and evaluation in wet-lab robotic applications.