通用学习新框架解决软体机器人控制难题

Core Insights - The research team from Southeast University, in collaboration with the National University of Singapore and MIT, has developed a universal learning and control framework for soft robots, inspired by the structure of brain neurons and synaptic plasticity [1][2]. Group 1: Framework Development - The proposed framework consists of two functional modules: one captures common features of soft robots performing various tasks using cameras, while the other adjusts control commands based on meta-learning gradient algorithms [1]. - The framework has been validated on three different types of soft robotic arms for trajectory tracking, object manipulation, and shape control [1]. Group 2: Performance Metrics - Experimental results indicate that the control framework maintains high precision and stability under complex conditions, such as varying loads and environmental disturbances [2]. - The absolute positioning control error is within 5 millimeters, and the deformation control accuracy exceeds 92% [2]. - Compared to mainstream Gaussian process-based control methods, the positioning control error is reduced by 44% to 55%, and the deformation control error is reduced by 33% to 68% compared to image-based inverse kinematics methods [2].