For mobile robot phototaxis control problems, the human or animal “perception-action” cognitive mechanism is simulated. The structure of mobile robot is designed and the method of phototaxis control is proposed based on the Boltzmann machine neural network. The Boltzmann machine neural network is trained by the knowledge set. The phototaxis control method is implemented by using the Boltzmann machine neural network operation mechanism. Simulation results show that the proposed method can improve the control accuracy and the success rate of robot learning.