TLDR: The DexHand 021 is a new lightweight (1kg) robotic hand with 19 degrees of freedom, inspired by human hand biomechanics. It features a cable-driven system with artificial muscles, advanced tactile and angle sensors, and a proprioceptive compliance control method. This control system, which uses joint torque estimation, reduces joint torques by over 31% during grasping compared to PID control, enhancing safety and efficiency. The hand excels in both power and precision grasps, capable of performing 33 GRASP taxonomy motions and complex manipulation tasks.
Robotic hands have long sought to mimic the incredible dexterity and adaptability of the human hand, a challenge that involves balancing complex mechanics, sensing capabilities, and coordinated movement. A new research paper introduces the DexHand 021, a significant step forward in this field, presenting a high-performance, cable-driven robotic hand designed with bio-inspired principles and advanced control systems.
The DexHand 021 stands out with its impressive specifications: a five-finger design boasting 12 active and 7 passive degrees of freedom (DOFs), achieving a total of 19 DOFs. This dexterity is packed into a lightweight 1kg design, making it highly agile. The hand demonstrates a single-finger load capacity exceeding 10 Newtons, fingertip repeatability under 0.001 meters, and remarkably low force estimation errors, below 0.2 Newtons. These capabilities are crucial for precise and robust manipulation in various environments.
One of the core innovations in the DexHand 021 is its bio-inspired design, drawing parallels from human hand biomechanics. It utilizes a tendon-driven system, where custom artificial muscle units simulate the contraction dynamics of biological muscles. These units are powered by high-power-density hollow-cup DC motors coupled with worm gear mechanisms, pulling multi-braided, high-strength tungsten cables as tendons. This design allows for human-like dexterity while addressing engineering constraints like size and weight.
The hand features a modular design for its fingers, including a distinct thumb module and a shared design for the other four fingers. This modularity contributes to its maintainability and cost-effectiveness. Each finger incorporates underactuated joints to reduce complexity while preserving essential functionality. For instance, the thumb module uses three motors to actuate four joints, while the four-finger module uses two motors to drive three joints, optimizing space and performance.
A sophisticated sensing system is integrated into the DexHand 021. Its fingertips are equipped with multi-point capacitive tactile sensors that can detect normal and tangential forces with high accuracy, and even sense the dielectric constant of different materials. Additionally, each actively controlled joint includes a Hall-effect angle sensor, providing precise motion tracking. This comprehensive sensing suite is vital for the hand’s ability to interact intelligently with its environment.
The control strategy for DexHand 021 is another highlight, featuring a proprioceptive force-sensing-based admittance control method. This advanced control system enhances manipulation by allowing the hand to adapt its stiffness and damping in response to external forces, preventing overload during collisions and improving grasping stability. It utilizes Gaussian Process Regression (GPR) for accurate joint torque estimation, which is critical for the admittance control to function effectively. Experimental results show that this method significantly reduces joint torques during multi-object grasping by 31.19% compared to traditional PID control, leading to reduced energy consumption and extended hardware lifespan.
The performance of the DexHand 021 has been rigorously validated through extensive experiments. It successfully executed 33 GRASP taxonomy motions, encompassing both power and precision grasps. The hand demonstrated its capability in complex manipulation tasks, such as using tweezers, cutting bananas, operating a pipette, writing, and even solving a Rubik’s Cube. These achievements underscore its potential for a wide range of applications, from industrial automation to medical rehabilitation and human-robot collaboration.
Also Read:
- Advancing Robot Dexterity: A Collaborative AI and VR Framework for Enhanced Manipulation
- Real-DRL: Bridging the Gap for Safe AI in Physical Systems
This work represents a notable advancement in the development of lightweight, industrial-grade dexterous robotic hands and the enhancement of proprioceptive control. The researchers envision future iterations integrating full tactile sensing and embodied intelligence to further expand its operational capabilities. For more in-depth technical details, you can refer to the full research paper here.
