Point-Cloud-Based Road Course Estimation on Automotive Radar Data

One of the most significant capabilities of an autonomous vehicle is planning its driving path based on the current surroundings. Radar can be of great assistance in this endeavor because of its proven ability to detect objects at a long range and work robustly in all kinds of weather conditions. Planning the road course is usually realized by an analytical methodology or a deep-learning based approach that relies on radar grid maps. In this paper, our new algorithm works directly on the noisy radar point cloud and achieves almost 5% improvement over the gridmap-based approach. The result shows that deep learning with point cloud data is also a promising way to achieve accurate road course estimation.