EquiDexFlow: Contact-Grounded SE(3)-Equivariant Dexterous Grasp Generative Flows

2026-06-10

AI Summary

This work addresses the physical instability in existing dexterous grasping methods, which typically model contact forces in a post-hoc manner. The authors propose the first end-to-end SE(3)-equivariant flow-matching model that jointly predicts wrist pose, joint angles, contact points, surface normals, and contact forces directly from object point clouds. By embedding geometric constraints into the architecture itself, the model ensures contacts lie strictly on the object surface and that contact forces adhere precisely to the Coulomb friction cone—achieving physically compliant grasps without relying on loss-based penalties. Trained on 81 objects, the method yields zero friction violations, achieves state-of-the-art composite scores, and minimizes torque residuals. In real-world experiments, a robot successfully performs open-loop stable grasps on all six test objects, including highly asymmetric ones under large rotational perturbations, and supports cross-hand retargeting and hardware fine-tuning.

Abstract

Most learned dexterous grasp generators relegate contact forces to a downstream verification step, so a kinematically-plausible pose can still violate the conditions for a stable physical grasp. We address this with EquiDexFlow, an SE(3)-equivariant flow-matching model that jointly predicts wrist pose, joint angles, fingertip contacts, surface normals, and contact forces from an object point cloud. Our architecture projects contacts onto the object surface and forces into the Coulomb friction cone by construction, so placement and friction compliance hold without loss penalties. We prove end-to-end SE(3) equivariance and verify it empirically over 200 rotations, with wrist residuals below $0.04^\circ$ and exactly zero joint deviation. Trained on 8,100 force-closure grasps across 81 objects for the 16-DoF Allegro Hand, our model achieves zero friction violations, the best composite score, and the lowest wrench residual among all ablation variants. We retarget decoded fingertip contacts to a 16-DoF LEAP Hand via per-finger inverse kinematics, and our hardware-feasible refinement places every joint at least 5% inside its actuator envelope while preserving wrench balance. On the physical robot, retargeted EquiDexFlow-decoded grasps complete open-loop pick-and-hold trials on all six test objects, with every asymmetric object succeeding at both the canonical pose and a $120^\circ$ co-rotation. Videos, code, and checkpoints are available at https://equidexflow.github.io.

Problem

• dexterous grasp
• contact forces
• force closure
• SE(3) equivariance
• friction compliance

Innovation

• SE(3)-equivariance
• dexterous grasp generation
• contact-grounded modeling
• friction cone compliance
• flow matching