DragMesh-2: Physically Plausible Dexterous Hand-Object Interaction with Articulated Objects

2026-06-13

AI Summary

This work addresses the challenge that dexterous hands struggle to reliably drive target parts of articulated objects through physical contact, and existing methods exhibit poor generalization under varying contact loads. The authors propose a contact-driven interaction framework that models articulated motion as the outcome of sustained contact between the hand and object handles, introducing PICA—a physics-aware training mechanism that operates without tactile or force feedback. This approach represents the first paradigm shift from object-centric trajectory generation to hand-contact-driven manipulation, relying solely on geometric information to construct interaction data. Experiments across seven categories of GAPartNet articulated objects demonstrate that the method achieves higher task success rates under diverse damping conditions and significantly improves robustness to variations in contact load compared to state-of-the-art approaches.

Abstract

Dexterous interaction with articulated objects is important for household, assistive, and humanoid manipulation, where multi-finger hands can provide compliant contact patterns beyond parallel-jaw grasping. However, articulated-object manipulation differs from static-object manipulation: the target part cannot be directly actuated, and its motion must emerge through sustained physical hand--handle contact. This makes the transition from object-centric articulated generation to hand-driven dexterous hand--object interaction non-trivial, since geometric trajectory replay or open-loop execution does not model the contact dynamics required to move the articulated part. Moreover, policies trained only for task completion under fixed dynamics can overfit nominal contact loads, especially without tactile or force feedback, and may degrade when the contact load changes. To address these challenges, we present DragMesh-2, a contact-driven framework for dexterous interaction with articulated objects that extends articulated interaction from object-centric generation to hand-driven dexterous hand--object interaction, where articulated motion must arise through physical contact. We further propose PICA, a physically informed contact-aware training mechanism that injects physical signals into policy learning without tactile or force feedback, improving robustness and task success under changing contact loads. Finally, we conduct systematic evaluation across multiple damping conditions and articulated-object categories to study robustness under contact-load variation, and provide a pure-geometry dexterous interaction resource to support future loco-manipulation and humanoid hand--object interaction research. Across seven GAPartNet objects, DragMesh-2 achieves stronger robustness under contact-load variation than the compared methods while maintaining high task success across damping conditions.

Problem

• dexterous manipulation
• articulated objects
• hand-object interaction
• contact dynamics
• physical plausibility

Innovation

• dexterous manipulation
• articulated objects
• contact-aware learning
• physical plausibility
• hand-object interaction