Visual Path Following and Obstacle Avoidance by Artificial Neural Networks

The aim of this work is to propose an artificial neural network based solution to a robot guidance problem, namely to visual path following and obstacle avoidance. Robot guidance task is performed through neural algorithms that allow a controller to determine robot's location within an environment like a building and to re-plan the trajectory of the robot both in the case that small deviations from the prescribed one are observed and in case of large obstacles to be avoided. The visual information only is exploited, and the neural networks are used both to learn the robot control laws and to extract from the visual data the necessary cues for obstacle avoidance. In this paper we also propose a path planning procedure in presence of unknown obstacles, which are detected by distance sensors. We suppose that the robot knows its position as well as the target one. The path planning procedure employs two different algorithm: the surrounding algorithm (SA) and the navigation algorithm (NA). The first one is activated when the robot is close to an obstacle, while the second one is useful to reduce the length of the path.