Chronological feed of everything captured from Dorsa Sadigh.
HandelBot addresses the sim-to-real transfer failure mode for millimeter-precision dexterous tasks by layering two adaptation stages on top of a simulation-trained policy: a structured refinement pass that corrects lateral finger joint alignments from physical rollouts, followed by residual reinforcement learning to autonomously acquire fine-grained corrective actions. The system achieves successful bimanual piano playing across five songs with only 30 minutes of real-world interaction data, outperforming direct sim deployment by 1.8×. This demonstrates that targeted, staged adaptation — rather than end-to-end real-world training — can unlock high-precision dexterous manipulation at practical data costs.