A Graph-search based Trajectory Planning Approach with Longitudianl Priority

Most optimizations for trajectory planning with autonomous vehicles first decouple the lateral-longitudinal-time problem, transforming decision-making into a matter of limited choices. Potential alternatives that can lead to better performance have been ruled out once the trajectory has been predefined in spatial dimensions, such as path optimization. This paper mainly draws on the optimization method from the Apollo EM Motion Planner and proposes a simplified, graph-search based approach by putting the decision-making process as well as the speed optimization first. The approach mainly concentrates on multilane scenarios with high-speed autonomous driving in a two-layer manner. The first layer is the combination of decision-making and longitudinal planning, whose purpose is to provide optional speed profiles under different target distances and local decisions. The second layer supplements the lateral planning process depending on every achievable speed profile. Dynamic programming is selected in both two layers to search the best discrete sampling path, because the optimization mode with longitudinal priority reduces pruning effort of dynamic programming in the second layer. Compared with the EM Motion Planner, this approach makes up the possible loss with different target distances and local decisions, it also simplifies the detection of obstacles. In the simulation, the capabilities of the approach are testified under a high-speed, multilane scenario with obstacle vehicles with random speed.