Using visual perception for controlling an outdoor robot in a crisis management scenario

Crisis management teams (e.g. fire and rescue services, anti-terrorist units ...) are often confronted with dramatic situations where critical decisions have to be made within hard time constraints. Therefore, they need correct information about what is happening on the crisis site. In this context, the View-Finder projects aims at developing robots which can assist the human crisis managers, by gathering data. This paper gives an overview of the development of such an outdoor robot. The presented robotic system is able to detect human victims at the incident site, by using vision-based human body shape detection. To increase the perceptual awareness of the human crisis managers, the robotic system is capable of reconstructing a 3D model of the environment, based on vision data. Also for navigation, the robot depends mostly on visual perception, as it combines a model-based navigation approach using geo-referenced positioning with stereo-based terrain traversability analysis for obstacle avoidance. The robot control scheme is embedded in a behavior-based robot control architecture, which integrates all the robot capabilities. This paper discusses all the above mentioned technologies.