An Intention-aware and Online Driving Style Estimation Based Personalized Autonomous Driving Strategy

Autonomous vehicles are aiming at improving driving safety and comfort. They need to perform socially accepted behaviors in complex urban scenarios including human-driven vehicles with uncertain intentions. What's more, understanding human drivers' driving styles that make the systems more human-like or personalized is the key to improve the system performance, in particular, the acceptance and adaption of autonomous vehicles to human passengers. In this study, a personalized intention-aware autonomous driving strategy is proposed. An online driving style identification is proposed based on double-level Multi-dimension Gaussian Hidden Markov Process (MGHMP) with arbitration mechanism and evaluated in field test. A Mixed Observable Markov Decision Process (MOMDP) is built to model the general personalized intention-aware framework. A human-like policy generation mechanism is used to generate the possible candidates to overcome the difficulty in solving MOMDP. The index of surrounding vehicles' intention of the upper-level MGHMP is updated during each prediction time step. The weighting factors of the reward function are configured with the identification result of lower-level MGHMP. The personalized intention-aware autonomous driving strategy is evaluated on a Real-Time Intelligent Simulation Platform. Results show that the proposed strategy can achieve the online identification accuracy above 95 % and for personalized autonomous driving in scenarios mixed with human-driven vehicles with uncertain intentions.