Kinodynamic Motion Planning Using Multi-Objective optimization

As autonomous driving gains importance, universally applicable motion planning approaches that offer safe and comfortable rides have to be developed. Most planning methods up-to-date still struggle when dealing with dynamic environments. They require extensive parameter-fine tuning in order to generate comfortable and safe solutions and it is not known prior to optimization which set of parameters would produce the 'best' solution. Therefore, we introduce a multi-objective optimization that plans a set of trajectories using several weights and targets (e.g. desired velocity or lanes). Thus, reducing the need of extensive parameter fine-tuning and increasing the planner's capabilities to handle dynamic environments. Furthermore, in order to plan multiple trajectories in real-time, a smart-initialization of the optimization problem is introduced that speeds up the multi-objective optimization further. Due to the proposed architecture that consists of a Planning-, Evaluation-and Selection-module, the planner is capable of providing a high level of comfort and safety - even in the case of non-convergence of the optimization. The novel motion planning approach is evaluated in terms of its applicability and performance.