Event
Survey reviews 113 learning-based dexterous manipulation methods
1 reports · 1 sources
Luai Abuelsamen published a survey reviewing 113 learning-based dexterous manipulation methods, covering the dexterous hands used, training simulators, and how results are reported. The author states that nearly half of the works follow the same pipeline: reinforcement learning on privileged state in a GPU simulator, then distillation into a vision-only student model and transfer to hardware; almost all of these works are concentrated on the same simulator, while later versions of that simulator are rarely used. The survey also notes that another, faster-growing line instead learns from teleoperated demonstrations, including vision-language-action models, diffusion policies, and systems whose main contribution is the data-collection rig itself. The author states that when the depth of finger indentation into the grasped object is reported, that depth is not measured in the physical environment that produced the action. Section 7 of the survey proposes a reporting protocol for these unreported quantities.
Report timeline
- Survey: Dexterous hands, simulators, training, and evaluation in 113 learning-based manipulation methods
A survey reviews 113 learning-based dexterous manipulation methods, covering the dexterous hands used, the training simulators, and how results are reported. The authors note that nearly half of the works use the same recipe: reinforcement learning on privileged state in a GPU simulator, then distillation into a vision-only student model transferred to hardware, and nearly all of them concentrate on the same simulator.
arXiv cs.RO ·