Probabilistic terrain modeling of mobile robots for outdoor applications using a scanning laser

In this paper the real-time three-dimensional terrain estimation and modeling problem of mobile robots for outdoor applications is considered. A uniform probabilistic sensor model for scanning lasers is created to transform measurements into elevation estimation in the form of normal distributions by multi-level coordination transformation, with uncertainties from un-modeled errors or sensors accuracies are treated as Gaussian noises. Under the elevation map description framework of the terrain model, the obtained elevation estimations are assigned to terrain cells according to their confidence boundary, and can be fused in a probabilistic manner to update the elevation map. Moreover, the terrain occlusion are considered and solved by local window detection method to improve the accuracy of the terrain model. Outdoor experimental results of a tracked robot with a scanning laser are given to demonstrate the effectiveness of the proposed method.