Uncalibrated vision-based mobile robot obstacle avoidance

Abstract

Obstacle avoidance is an important element of mobile robot navigation. Here the obstacle avoidance problem is solved utilizing visual feedback in an optimization-based control technique. This work is part of a larger investigation of uncalibrated vision-based control of unmodelled robotic systems. Specifically, this paper will discuss the control of a mobile robot utilizing a fixed camera. The mobile robot's workspace is within the field of view of the camera. Control is formalized as the minimization of an image-based objective function related to some desired robotic behavior. This function is minimized utilizing a dynamic quasi-Newton method utilizing dynamic recursive least squares Jacobian estimation. An advantage of using Jacobian estimation is that the system model is estimated online. This approach provides a generic control method that can be used to control a variety of robotic systems with little a priori information. In addition, no camera calibration is necessary, allowing system reconfigurations. For obstacle avoidance, a potential field approach is used to set up repulsive functions around obstacles. The design and implementation of an objective function that will navigate the robot to a goal point while avoiding obstacles is discussed.