Mobile robot navigation using non-contact sensors

Sammanfattning: This thesis concerns localization and control of mobile robots using range measurements from optronic sensors, in particular scanning time-of-flight lasers. The thesis covers the localization algorithms, on-line sensor calibration and tele-operation for semi-autonomous control of mobile robots. A real-world application of an autonomous mobile robot equipped with an arm for surface finishing operations is also developed. The Range Weighted Hough/Radon transform is introduced for robust detection of walls and extraction of geometric parameters. Thus a robot can automatically create a map of its environment and update it using an extended Kalman filter. The required association of observations with the map features uses a Bayesian classifier. New walls are added to the map as they are observed and classified as new. Navigation is performed with typical fluctuations of 1 cm and 1 degree (standard deviation) at speeds of 0.5 m/s. The navigation is robust even in cluttered environments and with several persons moving around. Sheet-of-light range cameras have also been tested for mobile robot navigation. A calibration algorithm for simultaneous calibration and localization integrates the calibration into the localization system. The only absolute reference needed by the calibration algorithm is the relative motion of the robot. The resulting system performs calibration during normal operation without lowering the localization performance. The algorithms above were applied to two applications: - A mobile robot capable of autonomously spray-plastering the walls and ceiling of a room. The prototype was successfully tested at construction sites. - A semi-autonomous robot tele-operated using tele-commands. The tele- commands include navigation along corridors, through open doors, following walking persons and automatic map generation of the remote site.