A dexterous hand can reach the right pose and still drop its object. Researchers using the LEAP hand treat that risk as a separate safety problem. They train in simulation without human demonstrations and add an adjustable residual filter to moderate risky actions.

Across 13.3 hours of real-hand testing, four objects and more than 4,800 unseen goals, the base policy reached 95.2% of target poses within 2 cm and 0.4 radians. The filter cut drops by 75% while retaining 91.5% target success. The trade-off is measurable, but performance on other hands and industrial duty cycles remains unproven.