Learning and control of cooperative behaviors of wearable robot using inverse differential game

摘要

Building an intelligent machine that can seamlessly work with the human is a grand challenge. Observing humanhuman cooperation can give us a big clue to find out intrinsic principles, such that we can apply similar techniques onto the robots to let them cooperate with the humans. During the past decades, Game Theory has been widely investigated by many research communities to better describe the group behaviors as well as the human-human cooperation strategies. However, the lacking of systematic tools to retrieve the realistic objective functions makes this method hardly being verified. In this paper, we formulate the Human-Robot Interactions (HRIs) as linear Differential Games. In order to retrieve the objective function of the human, we proposed a new computational paradigm - Inverse Differential Game to learn the cost function parameters so as to interpret human cooperative behaviors. The mathematical approaches to solve the inverse problem using Convex Optimization have been derived. The experimental simulation was carried out to evaluate the feasibility of our algorithm.