Human-inspired robot task learning from human teaching

The ability of a service or personal robot to learn new tasks from human teaching is important if it is to be multi-functioning and serve users a lifetime. Considering the vast variation of tasks, work environments, and nature of potential teachers or users who may not have knowledge in robotics, the problem of task teaching and learning can be difficult to achieve. Current methods of robot teaching and learning do not yet enable the robot to learn different types of tasks from the teaching by a general user. This paper presents a human-inspired method of robot task learning from human instructive hand-to-hand teaching. The method is novel in including an introduction of the complete task to the robot before task demonstration, a voting algorithm for segmenting the demonstrated task trajectory, and a Bayesian approach to assign partitioned trajectory segments to subtasks. Also, the proposed trajectory blending scheme can generate actual task paths in real-time to adapt learned tasks to new task setups.