We are researching transporter mobility robots for a new generation of personal transportation infrastructure. This paper describes our autonomous navigation technology which is adapted to crowded pedestrian streets, and evaluation results in Tsukuba Challenge 2010 using an experimental robot Sofara-T. In crowded outdoor environments, there are many moving objects such as pedestrians and bicycles. And there are complex shaped objects such as trees and bushes with branches and leaves. For autonomous navigation, it is required that localization, road terrain analysis (obstacle detection) and local path planning (obstacle avoidance) methods are adaptable to dynamic environments and complex shaped landmarks. Primarily, we developed a 3D-LIDAR using gimbal mechanism to measure 3D shapes of surrounding objects as landmarks and obstacles at rapid speed. In localization, we utilize static objects in upper regions than humans as landmarks in order not to lose robot position caused by occlusion in crowds. Moreover we developed a new localization method with 3D map matching adapted to complex shaped landmarks. Our method extracts bounding directed points of occupied space as landmarks, and matches current measured points to reference map points utilizing normal directions and a constraint of the gravity direction. For highly reliable road terrain analysis, we fuse results of bump detection with a 3D-LIDAR and a stereo camera using a binary bayes filter. Then our local path planning in variable velocity can avoid obstacles with safety and smoothness. In Tsukuba Challenge 2010, our robot Sofara-T robustly ran 1.8[km] 14 times in 15 experiments at maximum velocity 3.96[km/h].
