Executive Summary : | This proposal aims to develop a learning control algorithm inspired by human-motor-learning processes and apply it to autonomous robots. The robots will learn the optimal mechanical impedance for a given task through repeated physical interactions with the environment, without using force-based or impedance-based strategies. The algorithms and robot implementation will be developed through extensive theoretical, computational, and experimental work. The proposed work is motivated by the growing demand for robots for tasks involving physical interactions with humans or other robots, which require higher levels of sophistication in sensing and control of forces and mechanical impedance. Humans are adept at modulating mechanical impedance as they interact with the environment, and recent studies reveal that humans can learn optimal stiffness behavior without explicit strategies for stiffness. The proposed work is organized into five technical tasks: developing a learning control algorithm, conducting a simulation study with a 2R robot, building and testing a 2DOF "environment" robot, building and testing the main autonomous 2R robot with appropriate learning and control features, and validating and demonstrating the learning capabilities of the main robot with a reaching task with various force perturbations. Preliminary results show promise that a learning control algorithm combining three components can show desired features.
This work will forge new directions in learning control and robotics, and will be an important step in developing safe collaborative robots and self-learning robots. The experimental development may also lead to intellectual property. |