RobotFingerPrint: Unified Gripper Coordinate Space for Multi-Gripper Grasp Synthesis and Transfer

TL;DR

The paper presents the Unified Gripper Coordinate Space ($S^2$) to unify grasp representations across diverse grippers, enabling synthesis and transfer without retargeting. A CVAE predicts object-centric UGCS maps $(\lambda,\varphi)$ from object point clouds, producing dense correspondences that drive a differentiable grasp optimization pipeline. The method demonstrates strong performance in simulation and real-world experiments, including transfer between unseen grippers and human-to-robot demonstrations, highlighting improved grasp diversity and transferability. This approach offers a practical platform for cross-gripper grasp planning and learning-from-demonstration applications with robust generalization capabilities.

Abstract

We introduce a novel grasp representation named the Unified Gripper Coordinate Space (UGCS) for grasp synthesis and grasp transfer. Our representation leverages spherical coordinates to create a shared coordinate space across different robot grippers, enabling it to synthesize and transfer grasps for both novel objects and previously unseen grippers. The strength of this representation lies in the ability to map palm and fingers of a gripper and the unified coordinate space. Grasp synthesis is formulated as predicting the unified spherical coordinates on object surface points via a conditional variational autoencoder. The predicted unified gripper coordinates establish exact correspondences between the gripper and object points, which is used to optimize grasp pose and joint values. Grasp transfer is facilitated through the point-to-point correspondence between any two (potentially unseen) grippers and solved via a similar optimization. Extensive simulation and real-world experiments showcase the efficacy of the unified grasp representation for grasp synthesis in generating stable and diverse grasps. Similarly, we showcase real-world grasp transfer from human demonstrations across different objects.

Figures (11)

• Figure 1: Comparison between three different representations for grasp synthesis: (a) contact map; (b) sparse gripper keypoints; (c) our unified gripper coordinate space
• Figure 2: Max spheres across all the 12 different grippers including the gripper set from the MultiGripperGrasp casas2024multigrippergrasp dataset.
• Figure 3: Maximal Graspable Sphere: We test for multiple spheres with varying radii in parallel using Isaac Sim nvidia2023-isaac-sim
• Figure 4: Visualization of Unified Gripper Coordinate System on grippers from MultiDex li2023gendexgrasp and their corresponding sphere points. The colors indicate where each gripper's region is mapped on the sphere surface.
• Figure 5: Ground-truth coordinate map $\Phi_O$ on the object set with some sample grasps from the MultiDex li2023gendexgrasp dataset.