A bio-inspired solution for a local autonomous, reflex, obstacle avoiding behavior

The main goal of our research consists in finding a simple, straightforward online solution for obstacle avoiding problem encountered in mobile robots. The solution allows the robot to develop a local autonomous obstacle avoiding behavior every time when a higher-level motor command that is driving the robot (e.g. go forward/backward) put it in imminent danger to collide. The solution we proposed for a robot with 36 evenly distributed infrared (IR) sensors is a very simple one, based only on a minimal artificial neural network (ANN) trained with a backpropagation-like algorithm. Computationally cheap, the online learning algorithm we implemented proved to be very successful in both, static and dynamic clustered environment. The results reported here were obtained in MobotSim 1.0.03 - a configurable 2D simulator of differential drive mobile robots.