Morphometric Imitation: From Morphology and Contact Aware Hand Retargeting to Sim-to-Real Visuomotor Policy

Morphometric Imitation turns human hand-object demos into robot policies with 89.3% zero-shot success in 300 real trials.

Morphometric Imitation converts reconstructed human hand-object interactions into zero-shot sim-to-real visuomotor policies for dexterous robots. Morphometric optimization retargets motion across hand morphologies while preserving contacts, residual reinforcement learning makes the references dynamically feasible using object pose and contact cues, and the resulting demonstrations are distilled into vision-based policies. Across three robot hands and ten interactions, contact F1 improves by at least 8 points over the best of five baselines and downstream dynamic retargeting success by up to 35 points. The policies achieve 89.3% zero-shot success in 300 real-world trials on 30 objects.