A learning/adaptive robot controller

The authors explore the relationship of learning and adaptations in robot control. Learning, in this context, is the process of identifying the robot dynamics and its interaction with the environment for the purpose of improved tracking over an infinite horizon. Adaptation is the process of adjusting the controller to comply with the regulation and tracking needs of the closed-loop system. It is thus demonstrated that learning conflicts with adaptation in its tendency to increase the present tracking error, due to the minimization of different criteria (dual control principle). Exploratory schedules (ES) are reference trajectories which are specifically designed to provide efficient closed-loop learning. The authors relate ES design to the issue of input richness (or persistent excitation).<
>