Nonholonomic Robot Stabilization by Vision{based Time{varying Controls

Docking or parallel parking maneuvers of mobile manipulators with nonholonomic constraints, guided by vision sensory data, are considered. This corresponds to the stabilization of a highly nonlinear system to a desired pose, for which time{varying state feedback controls are employed, using a vision{based approximation of the system's state. The vision data are provided by a camera carried by the arm of the mobile manipulator, which tracks a target of reference, as the robot moves. Two approaches are considered: the rst involves continuous homogeneous time{varying controls, where the state information is updated at frame rate from vision. In order to improve robustness with respect to modeling errors, a second approach is explored, namely hybrid time{varying controls involving a combination of open{ and closed{loop phases, where the state information is updated from vision data only at the beginning of each period of the periodic open{loop controls. The experimental evaluation of the proposed techniques employs a nonholonomic mobile manipulator prototype developed in the robotics laboratory of INRIA Sophia{ Antipolis, which o ers the possibility of testing image understanding and control schemes in real{time on a dedicated multiprocessor system. The sensory apparatus of this robot is only crudely calibrated, hence our interest in control laws robust to modeling errors. Joint work with Claude Samson, Projet Icare, INRIA Sophia{Antipolis, France