Humanoid robot navigation by probabilistic multiple stereo matching

This paper proposes a visual navigation method for humanoid robot, based on the probabilistic multiple stereo matching. In this paper, we assumed that the existence of measured obstacle is probabilistic. We also consider that stereo vision has high reliability range for location estimation according to the distance between cameras. We therefore propose a stochastic model to measure obstacle location using successive parallactic images in different distance between cameras. We then implement a navigation system for humanoid robot, HOAP-1, using proposed model. This paper also reports performance results of autonomous walk to reach goal position avoiding several obstacles.