1 Mapping an outdoor environmnent for path planning. Maite López-Sánchez*, Gaurav Sukhatme and George A. Bekey. Robotics Research Laboratory University of Southern California (USC), USA. (*) Visitor scholar from the Institut d’Investigació en Intel.ligència Artificial (IIIA-CSIC), Spain. Abstract We present an incremental map building approach that is applied by a group of cooperating heterogeneous robots. Robots cooperate by sharing information in order to build their own maps. Environment information comes into the mapping process from two different sources: aerial images from a helicopter and sonar readings from several ground robots. Ground robots use the resulting map to plan paths towards goal positions. These paths avoid detected obstacles and are updated when there is new information about an obstacle obstructing them. Finally, we consider environment uncertainty depending on the reliability of the information. We use uncertainty as an estimation of the real existence of detected obstacles and we apply it in order to plan paths that may need to go trough non-real obstacles. 1. Introduction This research report is the result of working on a specific problem within a larger project. The problem is one of representing and managing information about an outdoor environment in a way that it is useful for path planning purposes. The generic goal of the longer project is to achieve cooperative tasks with a group of heterogeneous ground based and airborne vehicles. In the work presented herein, the different classes of vehicles accomplish cooperation by sharing information about the environment. We consider an outdoor environment which is unknown a priori. The global scenario of this project is a group of autonomous vehicles that consist of a helicopter and five ground robots that cover a certain area of an arena or reach a specific position specified as a command from a human. Our work has been focused in the representation of the environment as well as in the generation of paths that can be useful for the ground robots to avoid obstacles while reaching the goal position. Considering all the obstacle information that is available at each time, we obtain the shortest paths. However, we do not always use all the information details, therefore we can not guarantee optimality (in terms of distance) although we obtain optimal paths for each level of detail. Nevertheless the utility of a path strongly depends on the reliability of the information. In our approach we have two different sources of information: first, the helicopter has a camera facing the ground that provides a birds-eye view of the arena, and second, each ground robot has seven sonar sensors that detect ground obstacles. Robots communicate the obstacle information they gather in order to complete the map of the environment. Since every robot receives information about the obstacles detected by the rest of the robots, all of them have the same individual maps (or similar, in case of transmission problems). The distribution of information among the robots allows other robots to use information gathered by a robot that reached a specific region before them. Nevertheless, this does not prevent any robot to plan its own path under bad communication circumstances: since a robot keeps its own version of the map it can still plan a path albeit with less information. In the following section we describe how information from the different sources is preprocessed and used as input data for mapping. The third section gives the details of how our incremental