Shulou(Shulou.com)11/24 Report--

CTOnews.com, April 28, can robotic hands manipulate all kinds of objects as flexibly as humans? Researchers at Columbia University have recently made a breakthrough by developing a highly dexterous robot hand that can not only sense touch, but also master complex manipulation skills through machine learning algorithms.

The robotic hand has five fingers, each equipped with advanced optical tactile sensors that can accurately locate contact points and cover complex surfaces. Each finger has three separate joints, which can be flexibly adjusted. In order to test the dexterity of the robot hand, the researchers chose a difficult manipulation task: to make the robot hand rotate the object at will while keeping the object stable. This task requires constantly repositioning some fingers, while others fix the object.

Surprisingly, the robot hand does not rely on any visual feedback, only based on the data of the tactile sensor and its own position to accomplish this task. This means that the robot hand can operate in very poor light conditions, even in the dark. The researchers say this tactile dexterity will open up new real-world applications for automation, such as logistics, material handling, advanced manufacturing and assembly.

A dexterous robot hand with five tactile fingers in order for the robot hand to learn this manipulation skill, the researchers used a method called deep reinforcement learning, which allows robots to learn new physical tasks through practice. Researchers have also developed some new algorithms to effectively explore possible manipulation strategies. They used the simulator as a training ground and completed the equivalent of a year's practice in a few hours. Then they transferred the skills trained in the simulator to the real machine and successfully achieved the dexterity they expected.

"although we are demonstrating a proof-of-concept task, we believe this dexterity will open up entirely new possibilities for automation in the real world," said Professor Mattejocali of Columbia University. some of the more direct uses may be in logistics and material handling to help alleviate the supply chain problems that have plagued our economy in recent years and to carry out advanced manufacturing and assembly in factories. "

Professor Giocali said that the ultimate goal in the robotics field is to achieve auxiliary robots in the home, which is a real test of dexterity. "in this study, we have shown that robotic hands can also achieve a high degree of dexterity based solely on the sense of touch," he said. once we combine visual feedback with touch, we hope to be able to achieve higher dexterity and one day be close to replicating human hands. "

Professor Giocali points out that a physical robot that is useful in the real world needs both abstract semantic intelligence (CTOnews.com) and embodied intelligence (the ability to interact with the world physically). Large language models such as OpenAI's GPT-4 or Google's PALM are designed to provide the former, while manipulative dexterity like this study represents progress for the latter. For example, when asked how to make a sandwich, ChatGPT will type a step-by-step list in the response, but to actually make a sandwich, you need a clever robot. Similarly, researchers hope that physically skilled robots can bring semantic intelligence out of the purely virtual world of the Internet for real-world physical tasks, perhaps even in our homes.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.