Behavior and Interaction-aware Motion Planning for Autonomous Driving Vehicles based on Hierarchical Intention and Motion Prediction

Safe motion planning in complex and interactive environments is one of the major challenges for developing autonomous vehicles. In this paper, we propose an interaction and behavior-aware motion planning framework based on joint predictions of intentions and motions of surrounding vehicles. A multimodal hierarchical Inverse Reinforcement Learning (IRL) framework is designed to learn joint driving pattern-intention-motion models from real-world interactive driving trajectory data. Using the learned models, the proposed approach can probabilistically predict the continuous motions partitioned by discrete driving styles and intentions, while taking into account of the interactions between participants' actions. A chance-constrained POMDP strategy is utilized to generate risk-bounded motion plans based on those predictions. Simulation case studies based on challenging scenarios demonstrate that the proposed framework is effective in predicting surrounding vehicles' behaviors and generating safe motion plans in complex traffic environments.