Dynamics of a learning controller for surface tracking robots onunknown surfaces

An extended Kalman filter is applied to simulated sensor information as an approach to the surface estimation problem. It is assumed that a robotic probe equipped with a tactile sensor is given the task of working with a completely unknown surface. Kinematics and control based on tactile measurements are briefly discussed. An estimator which provides surface information as obtained by an inherently noisy force sensor is designed. From these estimates, a controller is given the capability of learning the constraint surface, thereby rejecting the noisy sensor data. After a short time, surface tracking is similar to the case of constrained motion on known surfaces