Acquisition of Smooth, Continuous Obstacle Avoidance in Mobile Robots

To enable robots to cope with unforeseen situations, as well as the noise and variation inherent in the real world, we believe it is essential to incorporate plasticity into robot controllers, and use as little predeened knowledge as possible. We present rst results of experiments in autonomous competence acquisition with \FortyTwo", a mobile robot equipped with tactile, sonar, infrared and visual sensors, used at the Department of Computer Science at Manchester. In these experiments, associations between incoming sensor signals and corresponding motor responses are stored in an artiicial neural network, the learning process is independent from human interaction and, due to the type of network used, fast. Contrary to our earlier work, in which active output nodes of the network triggered compound motor actions, FortyTwo uses just two output nodes which function similar to motor neurons in animate systems: each node directly drives one of two drive motors. This means that a continous spectrum of motor actions can be acquired, resulting in a smooth and continous obstacle-avoiding motion of the robot.