Learning Generalizable Hand-Object Tracking from Synthetic Demonstrations

TL;DR

This paper introduces a synthetic-data-based system for learning generalizable hand-object tracking controllers, enabling dexterous manipulation without human demonstrations and extending to diverse objects and hand types.

Contribution

The paper presents HOP and HOT, novel components for synthesizing trajectories and bridging synthetic-to-physical transfer, advancing manipulation control learning from synthetic data.

Findings

Enables tracking of complex, long-horizon sequences

Works across diverse object shapes and hand morphologies

Achieves scalable manipulation controllers from synthetic data

Abstract

We present a system for learning generalizable hand-object tracking controllers purely from synthetic data, without requiring any human demonstrations. Our approach makes two key contributions: (1) HOP, a Hand-Object Planner, which can synthesize diverse hand-object trajectories; and (2) HOT, a Hand-Object Tracker that bridges synthetic-to-physical transfer through reinforcement learning and interaction imitation learning, delivering a generalizable controller conditioned on target hand-object states. Our method extends to diverse object shapes and hand morphologies. Through extensive evaluations, we show that our approach enables dexterous hands to track challenging, long-horizon sequences including object re-arrangement and agile in-hand reorientation. These results represent a significant step toward scalable foundation controllers for manipulation that can learn entirely from synthetic data, breaking the data bottleneck that has long constrained progress in dexterous manipulation.