Online Intention Prediction and Task Planning for Human-Robot Collaboration and Competition

Abstract:
Recognizing human intentions is important and critical for various human-robot collaboration/competition applications. In this paper, a method of predicting human intentions or actions, when a human is executing an action, is proposed. Robot observes the states of a system/environment, which are associated with the definitions of actions. A Bayesian model is created to describe the probability of possible actions of human during the execution of actions. The candidate intention of human will be continuously updated and be used as the basis for continuous task planning for robot. Planned actions or strategies for robot in a human-robot collaboration/competition task are switched, and modified using a cognitive control framework. Our algorithm estimates the intention of human and generate corresponding actions of robot before the execution of an action is done, which largely increase the feasibility of creating real-time collaborating robot for complex human-robot collaboration and/or competition applications. We evaluated our method on combined simulation and experiment platforms.