ContactOpt: Optimizing Contact to Improve Grasps

TL;DR

ContactOpt introduces a deep learning and optimization framework that enhances hand-object grasp poses by optimizing contact points, leading to more realistic and human-preferred grasps in image-based hand pose inference.

Contribution

It presents a novel differentiable contact model and an optimization method that improves hand pose accuracy by optimizing contact, incorporating mesh interpenetration to simulate soft tissue effects.

Findings

Grasps better match ground truth contact points.

Achieves lower kinematic error in hand poses.

Human participants prefer grasps generated by ContactOpt.

Abstract

Physical contact between hands and objects plays a critical role in human grasps. We show that optimizing the pose of a hand to achieve expected contact with an object can improve hand poses inferred via image-based methods. Given a hand mesh and an object mesh, a deep model trained on ground truth contact data infers desirable contact across the surfaces of the meshes. Then, ContactOpt efficiently optimizes the pose of the hand to achieve desirable contact using a differentiable contact model. Notably, our contact model encourages mesh interpenetration to approximate deformable soft tissue in the hand. In our evaluations, our methods result in grasps that better match ground truth contact, have lower kinematic error, and are significantly preferred by human participants. Code and models are available online.