Bearings-only path following with a vision-based potential field

In this paper, we present a vision-based path following algorithm for a non-holonomic wheeled platform. The algorithm is based on choosing control actions that minimise the value of a potential field cost function calculated directly from the image plane. The algorithm is suitable for teach and replay or leader follower implementations where the desired path is represented as a collection of images. The algorithm computes the cost function based on the relative bearings of features matched between the current and previously observed images. A forward prediction step is then used to determine the control action that will lead to the greatest reduction in the cost function. The algorithm is demonstrated on a 400 m path in an outdoor environment where the accuracy is shown to be similar to that of differential GPS.