Title:Adaptive tracking control for task-based robot trajectory planning

Abstract:This paper presents a -- Learning from Demonstration -- method to perform robot movement trajectories that can be defined as you go. This way unstructured tasks can be performed, without the need to know exactly all the tasks and start and end positions beforehand. The long-term goal is for children with disabilities to be able to control a robot to manipulate toys in a play environment, and for a helper to demonstrate the desired trajectories as the play tasks change. A relatively inexpensive 3-DOF haptic device made by Novint is used to perform tasks where trajectories of the end-effector are demonstrated and reproduced. Under the condition where the end-effector carries different loads, conventional control systems possess the potential issue that they cannot compensate for the load variation effect. Adaptive tracking control can handle the above issue. Using the Lyapunov stability theory, a set of update laws are derived to give closed-loop stability with proper tracking performance.