A Hand Telerobotic System with Enhanced Dexterity towards Dexterous Manipulation Data Acquisition

Ruize Wang, Yangbin Bao, Qi Ye, Peng Cheng, Jiming Chen, Gaofeng Li

Keywords: Hand Telerobotic System, Full State Motion Capture, Adaptive Hybrid Mapping, Dexterous Manipulation

TL;DR: We propose a hand telerobotic system with a highly under-actuated hand exoskeleton for precious full state human motion capture and an adaptive hybrid mapping method for mapping motions to Allegro Hand towards dexterous manipulation data acquisition.

Abstract: Although the multi-fingered hand brings more possibilities for human-like dexterous manipulation, it also introduces serious challenges in interacting with the dynamic and uncertain physical environments. Compared with two-fingered parallel grippers, the multi-fingered hands have more joints and diverse interaction postures. This makes the operation no longer limited in the one-dimensional opening/closing degree, but can realize flexion/extension, adduction/abduction, and rotating in three-dimensional space, which are more dexterous and closer to human manipulation. However, how to precisely capture these dexterous motions and map them to a robotic hand is extremely challenging for existing technologies. In this abstract, we propose a hand telerobotic system with a highly under-actuated hand exoskeleton and an Allegro Hand towards dexterous manipulation data acquisition. Meanwhile, we design a joint mapping method with joint-space reconstruction and a Cartesian mapping method with an auxiliary frame to ensure the human hand motion can be correctly and precisely mapped to the robotic hand. To enhance system adaptability in complex tasks, we further design a hybrid mapping method combining the Joint-Cartesian space to improve the adaptability of our system. Compared with existing works, our system is able to capture the full states of humans hands and replicate them on the robotic hand, especially for the adduction/abduction motions. Three dexterous manipulation tasks with grasping, rotating, screwing and tapping actions are designed to validate the dexterity of our system.

Submission Number: 16