MultiHand: Design and Verification of a Dexterous Hand with Multi-Modal Grasping Capabilities
Yaopeng Tian, Changqing Guo, Shoujie Li, Chenxin Liang, Junbo Tan, Xueqian Wang
AI summary
Problem
Traditional single-function robotic grippers struggle to handle the wide variety of shapes, materials, and weights encountered in everyday environments, limiting their practical utility.
Approach
MultiHand combines three specialized fingertips—a controllable electromagnet, a miniaturized suction array, and a heat-cool adhesive layer—with a jamming palm to switch grasping modes on demand.
Key results
- Over 90% grasping success rate across 102 diverse household and industrial objects
- Development of the Generic Scene Dataset (GSED) with 3D-scanned models and open-source assets
- Magnetic finger lifts up to 500g metal items, suction finger secures smooth/curved surfaces, and adhesive finger achieves 95% success on mesh and fabric
- Demonstrated complex dexterous operations including bottle opening, drilling, and barcode scanning
Why it matters
Provides a versatile, single-end-effector solution for robots to adapt to diverse real-world objects without hardware swaps, accelerating deployment in domestic and industrial settings.
Abstract
As the end-effector of a robot, the dexterous hand is responsible for grasping and manipulation tasks. With the expansion of application scenarios, a single grasping mode is no longer sufficient to handle different objects and complex environments. In this paper, we design a multi-modal grasping three-fingered dexterous hand, where each finger has a specific function. In addition to performing routine tasks, it is also capable of easily grasping thin and lightweight objects. We pro- pose and validate a dataset of over a hundred objects, covering categories such as food, sports equipment, and household items. The thumb of the dexterous hand in this design is a controllable electromagnet, the index finger is a suction cup finger, and the middle finger is a controllable adhesive finger, with grasping and releasing controlled through heating and cooling. The palm features a jamming design, enabling both grasping and increas- ing gripping friction. To validate the grasping capabilities of the dexterous hand, we conducted single-finger performance tests and successfully grasped over a hundred common objects, with a grasping success rate exceeding 90%. This dataset was created through 3D scanning, including 102 common objects, and some of the models are open source. The dataset url is: https://github.com/TNPoppy/MultiHand.