Advancements in Robotics: Achieving True Dexterity in Robot Hands
When you think about how your hands effortlessly perform tasks like operating a TV remote or using cutlery at a restaurant, it’s clear that our hands are incredibly skilled and sensitive. Robotics researchers have long been trying to replicate this level of dexterity in robot hands, but it has proven to be a challenging endeavor. However, thanks to recent advances in sensing technology and machine-learning techniques, the field of robotic manipulation is rapidly evolving.
A Breakthrough: Highly Dexterous Robot Hand
Researchers at Columbia Engineering have developed a highly dexterous robot hand that combines advanced touch sensing with motor learning algorithms. To showcase its capabilities, the hand successfully performed a challenging task: rotating an unevenly shaped object while maintaining a stable grip. Remarkably, the hand accomplished this without any visual feedback, relying solely on touch sensing. Furthermore, it operated without external cameras, making it immune to issues like lighting and occlusion. This means that the hand can even function in the dark or under challenging lighting conditions.
Leveraging Optics-Based Tactile Fingers
Prior to this breakthrough, the researchers collaborated on developing optics-based tactile robot fingers. These innovative fingers allowed for precise contact localization on complex surfaces and easy integration into robot hands.
Teaching the Hand Complex Tasks
To enable the hand to perform complex manipulation tasks, the researchers designed a robot hand with five fingers and 15 independently actuated joints. Each finger was equipped with touch-sensing technology. They employed deep reinforcement learning, combined with new algorithms for effective exploration of motor strategies, to teach the hand complex tasks. The learning process, done without any vision input, was carried out in simulation, where the robot hand completed approximately one year of practice in a few hours of real-time. This training was then transferred to the physical robot hand, resulting in the desired level of dexterity.
The researchers believe that this breakthrough opens up new possibilities for robotic manipulation in the real world. Applications could include logistics and material handling, advanced manufacturing, and assembly in factories.
Joining Abstract and Embodied Intelligence
Achieving true dexterity in robot hands represents a significant advancement in the field of embodied intelligence. While large language models offer abstract intelligence, physical robots with dexterity can bridge the gap between virtual knowledge and real-world tasks. The ultimate goal is to combine abstract intelligence with embodied intelligence to create robots that are truly useful in our daily lives, capable of performing tasks based on semantic understanding.
This groundbreaking research will be presented at the upcoming Robotics: Science and Systems Conference. The paper is also available as a preprint and offers a video demonstration of the highly dexterous robot hand in action.
[VIDEO: Watch the highly dexterous robot hand in action: https://youtu.be/mYlc_OWgkyI]