3D-vision-based robot navigation: first steps Luc Robert, Rfigis Vaillant and Michel Schmitt INRIA Sophia-Antipolis 2004 Route des Lucioles 06565 Valbonne Cedex France Abstract This article shows a way of using a stereo vision system as a logical sensor to perform mobile robot navigation tasks such as obstacle avoidance. We describe our system, from which the implementation of a task described by an automaton can be done very easily. Then we show an example of a navigation task. 1 Introduction Many approaches have been proposed to the problem of mobile robot navigation. Some aim at following trajectories by servoing on feature that might be detected and tracked by sensors (the edges of a road [7], predefined beacons [1]). Our main goal is to obtain a passive-vision-based system that can today navigate in an unknown environment, avoid obstacles and collect some information without being constantly looked after. To achieve this, we use a trinocular stereovision system, which is a powerful and reliable vision sensor device. This system avoids developing sophisticated algorithms to detect obstacles in 2D images. 2 Hardware and software tools 2.1 Description Our mobile robot is a robuter (fig 1), with two driving wheels and a front turret that supports the set of cameras and allows rotation around a vertical axis. The primitives we use to describe a robot move are directly correlated to the amount of rotation of each driving wheel to achieve this move. Actually, the low-level implementation invokes a P.I.D. supplied by an odometric sensor on each driving wheel. Computation is processed on the Capitan parallel machine and its host machine, a Sun workstation. The communication between these computers and the robot is established through a video and radio links.