Visual Based Localization of a Legged Robot with a Topological Representation

Abstract

In this chapter we have presented the performance of a localization method of legged AIBO robots in not-engineered environments, using vision as an active input sensor. This method is based on classic markovian approach but it has not been previously used with legged robots in indoor office environments. We have shown that the robot is able to localize itself in real time even inenvironments with noise produced by the human activity in a real office. It deals with uncertainty in its actions and uses perceived natural landmarks of the environment as the main sensor input. We have also shown that data obtained from sensors, mainly the camera, is discriminate enough and allows a fast convergence from an initial unknown state, in which the belief has been distributed uniformly over the set of states . We have also shown that the robot can overcome action failures while localizing, and it recovers from them in an efficient way. The set of observations we have chosen have been descriptive enough to be efficient in the localization process. We think that the way we determine the number of doors and ceiling lights has been the key for the success of the localization system. This approach depends o