Bimanual manipulation with dexterous hands is drawing increasing interest in robotics, but researchers have lacked a common testing ground. A new arXiv preprint introduces BiDex-100, a comprehensive benchmark spanning more than 100 simulation tasks specifically for this domain. The authors note that few existing studies have developed benchmarks tailored to dexterous manipulation, leaving a gap that BiDex-100 aims to fill.

The benchmark is intended for robotic action models, providing a standardized set of tasks that require coordinated use of two dexterous hands. By covering a wide range of scenarios, BiDex-100 offers a way to compare approaches and measure progress in a field that has relied on ad hoc or single-task evaluations. The abstract does not detail the exact task categories or evaluation metrics, but the scale and focus on bimanual dexterity are clear.

As a simulation-based benchmark, BiDex-100 can be used for training and testing without physical hardware, which may accelerate research. The preprint is available on arXiv under identifier 2610.25459, though the full methodology and task list are not included in the abstract. Researchers in robotic manipulation will likely find this a useful resource for benchmarking future models.