Motion planning based on simultaneous perturbation stochastic approximation for mobile auditory robots

In this paper, a motion planning method for mobile auditory robots is proposed based on an optimization technique. Since it is one of the most important abilities for auditory robots to recognize vocal messages correctly, the proposed method is designed to maximize the confidence measure of a speech recognition since the measure is thought to be strongly related to the accuracy of the speech recognition. However, the cost function to optimize is hard to model explicitly, and it is difficult to obtain the gradient that is normally utilized to derive the motion. In order to overcome this difficulty, simultaneous perturbation stochastic approximation(SPSA) that does not require an explicit model of the cost function is applied to generate robot motion. The effectiveness of the approach was verified through real experiments: the robot could get better speech recognition rate after it approached the sound source by measuring the confidence measure.