Acquisition of Direction Control Knowledge using Neural Network for Small-Diameter Tunnelling Robot.

This paper describes a method for acquiring direction control knowledge using a neural network for a small-diameter tunnelling robot. This method uses a neural network whose input is both the deviaition and angular deviation between the main tunnelling body and the designed line. The output is the head angle. The neural network learns from learning data which gives the relationship between [deviation, angular deviation] and [head angle] for a given initial deviation, angular deviation and soil condition. These date are obtained by a good direction control law. We can control the direction of the tunnelling robot by using the neural network after learing for any initial deviation, angular deviation and soil condition. We could acquire direction control knowledge for both feedback control and fuzzy control by using this method.