In this paper, an EKF-based robot self-localization is developed for a small-size humanoid robot navigating in environments with natural features. A novel algorithm for feature initialization is proposed for the EKF localization algorithm: the beacon positions are firstly obtained by extracting natural and apparent features from the image taken by a monocular vision. Then the "bad" feature points are detected and abandoned before they are chosen to be a reference beacon. Meanwhile, each obstacle in the environment is detected by gathering together a set of nearby features. A regional feature detection method is also employed to detect the obstacle for comparison. The developed algorithm is implemented on a PC-based controller for a small-size humanoid robot which has dimension of 40cm in height and 3kg in weight. Experimental results show that the proposed self-localization and obstacle detection algorithms have good performance for the humanoid robot navigating in the environment with natural features.
Relation:
Proceedings of the 2008 CACS International Automatic Control Conference，6pages
