Autonomous Robot Positioning by Visual Servoing

This paper describes a vision-based system for the autonomous positioning of a mobile robot. Both the pose estimation method and the robot control system are presented. The former finds the position and attitude of the camera mounted on the robot with respect to a known 3D target using a single image. The method uses geometric considerations which combine perspective projection and scaled orthographic projection of the extracted feature points. Once the robot position has been estimated, the movement required for reaching the desired position can be evaluated. The control of the robot is performed in a closed-loop to increase the precision and robustness of the system. At an early stage the pose estimation method has been validated using synthetic data generated using a pin-hole model for the camera. Then it has been tested in a practical environment (inspired by the playground for the RoboCup) using a Khepera robot. The system is able, starting from arbitrary positions in the environment, to extract and identify feature points, to calculate the current position of the vehicle and reach, after few iterations in closed-loop, the desired goal position, in spite of the significant lens distortion of the uncalibrated camera mounted on the Khepera.