Abstract: The accuracy and stability of position data, which is offered by the unique visual sensor, can not satisfy the requirement of control system. The learning function of the BP neutral network is used to solve this problem. The state information of football and robot is treated as sample data, the visual sensor, accelerometer and compass are treated as the input of network, and the outputs of the neutral network assist soccer robot to track football. Simulation shows that the neutral network improves the accuracy of the location for football and robot. The expected result is obtained. The average relative error is superior to the traditional Kalman filter.

Key words: football robot, sensor, data fusion, neutral network