Online and Incremental Contextual Task Learning and Recognition for Sharing Autonomy to Assist Mobile Robot Teleoperation

This contribution proposes a fast online approach to learn and recognize the contextual tasks incrementally, with the aim of assisting mobile robot teleoperation by efficiently facilitating autonomy sharing, which improves our previous approach, where a batch mode was adopted to obtain the model for task recognition. We employ a fast online Gaussian Mixture Regression (GMR) model combined with a recursive Bayesian filter (RBF) to infer the most probable contextual task the human operator executes across multiple candidate targets, which is capable of incorporating demonstrations incrementally. The overall system is evaluated with a set of tests in a cluttered indoor scenario and shows good performance.