On-road Trajectory Planning with Spatio-temporal Informed RRT

In this study, a 3D sample trajectory planning approach for on-road scenarios is provided, which can directly supply the coordinate and velocity of each planning point. Simultaneously, increase sample pertinence and improve sampling efficiency. To generate a coarse trajectory in the Cartesian-time coordinate system, we utilize a method based on Informed RRT*. However, owing to the random nature of the sample-based method, the quality of the solution is dependent on the density of the sample. When the sampling density setting is low, the solution is not practicable or has no solution, and when the setting is high, dimension curse may occur. A function to identify barriers ahead of the vehicle is implemented before sampling to mitigate the impact of this flaw on online planning in extreme situations such as cornering and U-turn. We also sample the endpoint and circle the whole sampling region before planning for this task. As a result, the algorithm becomes more targeted. We used B-spline to refine the initial trajectory after acquiring it. The reliability of this method is demonstrated by a real-world vehicle experiment.