Mobile robot navigation using a neural net

For mobile robot navigation in an unknown and changing environment, a reactive approach is both simple to implement and fast in response. A neural net can be trained to exhibit such a behaviour. The advantage is that, it relates the desired motion directly to the sensor inputs, obviating the need of modeling and planning. In this work, a feedforward neural net is trained to output reactive motion, in response to sonar range inputs, with data generated artificially on the computer screen. The authors develop input and output representations appropriate to this problem. A reactive robot, being totally insensitive to context, often gets trapped in oscillations in front of a wide object. To overcome this problem, the authors introduce a notion of memory into the network by including context units at the input layer. The authors discuss the mode of training for such a network and present simulated runs of the trained net in various situations. The authors conclude by demonstrating a short tour of an actual mobile robot in their laboratory under the control of the trained neural net.