Design of the multiple Neural Network compensator for a billiard robot

Abstract

In this paper, a multiple Neural Network (NN) compensator is designed for a billiard robot to finish a task, in which the trained robot is commanded to control the cue ball to a specific target point along a trajectory with multiple cushion rebounds. A novel pyramid classification has been established to sort out the pattern of trajectory and its segments. For each trajectory pattern, a corresponding Back Propagation Neural Network (BPNN) model has been established to fit the deviation between theoretical direction point and actual one. The pyramid classification and a finite number of BPNN models composited the multiple NN compensator. In the test, the robot will calculate the deviation and work out the actual direction point for potting. The test results have verified the reliability and workability of the multiple NN compensator.