PURPOSE: To allow a robot to easily acquire direction control knowledge by letting the robot learn while using a neural network applying the deflection and deflection angle of a robot body relative to a planned line to input and head angles to output.
CONSTITUTION: A neural network 1 comprising an input layer 2, intermediate layer 3 made of at least one layer, and output layer 4 is constituted in such a manner that the deflection (a) and deflection angle (b) of a small bore robot body relative to a planned line are input and the head angle (c) is output from the output layer 4. A certain specified initial deflection acquired from the direction control rule of a specified small bore tunnel robot, deflection to deflection angle and soil property, and the relationship between the deflection and head angle are used as learning data, and a robot is made to learn by the learning rule of the neural network. With this constitution, a neural network type direction controlling rule which can cope with any initial deflection, deflection angle and soil property can be provided.
TAKEDA KOKI
YABUTA TETSUO
