For autonomous navigation applications, real-time modeling of the surrounding environment is a key component. Many approaches focus on how to interpret exteroceptive sensor data to obstacle information, with well-tailored sensor models to enable the transform and tackle the sensor uncertainties. However, sole information of obstacles is far from adequate for autonomous perception. The ego vehicle needs semantic information to understand the accessibility of the space. In the literature, methods have been proposed to encode prior map information into grids, the co-called semantic grid maps offer a new method to represent prior information. In this article, we propose a grid-based evidential approach to fuse exteroceptive sensor information with prior map knowledge. Each source of information is transformed into a specially designed evidential grid, the combination is carried out layer-wise based on the belief function theory. We report real results carried on public roads with the real-time software.

CICTP