DexMimic: A Generalist Dexterous
Hand–Object Tracking Policy
via Value-Guided Reset Allocation

Appendix
0:00 / 2:52

Abstract

Achieving human-like dexterous manipulation remains a central challenge in robotics. Learning from human hand–object interaction (HOI) datasets offers a promising pathway to diverse dexterous skills, yet existing physics-based approaches train specialist policies for individual trajectories, thereby limiting skill reuse at scale. We introduce DexMimic, a generalist dexterous hand–object tracking policy jointly trained across diverse HOI trajectories. Efficient joint training requires allocating each rollout to both an undertrained phase and a starting state that supports continued tracking. DexMimic realizes this insight through Value-Guided Reset Allocation, which adapts the initialization distribution online to prioritize difficult phases and select reset states with higher expected tracking returns.

DexMimic overview: human demonstrations, generalist policy learning, and dexterous hand–object manipulation results.

Method Overview

Value-Guided Reset Allocation: a reset map, phase and state allocation, and physics rollouts that jointly update the policy, critic, and initialization distribution.
Value-Guided Reset Allocation. The reset map stores candidate states and their critic values for each trajectory and reference phase. As training progresses, reset phase allocation prioritizes under-learned phases, while reset state allocation favors candidates with higher expected tracking returns within each selected phase. Physics rollouts from the sampled initializations update the policy and critic, replenish the reset map, and refine phase sampling probabilities, forming an adaptive training loop.