MACHINE PERCEPTION AND ROBOTICS GROUP

Dept. of Robotics Science and Technology,
Chubu University

Digital maps are data that numerically represent map information necessary for autonomous driving. The digital map is used for estimating vehicle positions, surrounding environments, and determining moving paths to a destination. Because it includes important information such as surrounding environments and road signs. Path information is also included for safe autonomous driving by following traffic rules. However, such path data is annotated manually and it is costly. Therefore, we aim to create path-planning data automatically on digital maps. Then, we propose a path-planning approach for vehicle movement. Our approach defines a cost function based on semantic scene labels and creates a minimum and optimal path. To estimate such a path, we use optimal rapidly-exploring random trees. Thus, it is possible to estimate an optimal path efficiently. Optimal cost is determined by learning with vehicle moving-path data. We also made a dataset for creating paths at intersections for quantitative evaluation. The results indicate that our proposed approach can create a optimal moving path.