Researchers at MIT have developed a control algorithm that allows a shape-shifting robot to autonomously move, stretch, and shape itself to complete tasks. The algorithm uses reinforcement learning and a coarse-to-fine methodology to control groups of muscles and optimize action plans. A simulation environment called DittoGym was created to test the algorithm, and it outperformed other methods in completing shape-changing tasks. This research could pave the way for the development of versatile, adaptable robots that can change their shape to accomplish various tasks.
18 Impressions