Brief

Researchers have introduced DragMesh-2, a new framework designed to improve dexterous interaction with articulated objects, such as doors or levers, which require sustained physical contact to manipulate. This framework extends previous object-centric generation methods to focus on hand-driven interaction, ensuring that articulated motion emerges through contact dynamics. To enhance robustness against varying contact loads without tactile feedback, the team developed PICA, a physically informed training mechanism. Evaluations on seven GAPartNet objects demonstrated that DragMesh-2 offers superior robustness and task success compared to existing methods. AI