A lane-changing behavioral preferences learning agent with its applications

Traditional lane-changing (LC) behavioral researches usually focus on the driver’s cognitive performance which includes the driver’s psychological and behavioral habit characteristics, rarely involving the affection of expert driver’s comprehensive behavioral preferences, such as: safety and comfort performance in LC process. Towards the free LC process, a novel LC safety and comfort degree index is proposed in this paper, as well as, the novel definition of LC driving behavioral preferences is described in detail. Taking advantage of interactive evolutionary computing (IEC) and real-time optimization (RTO) metrics, a kind of LC behavioral preferences on-line learning agent extending traditional Belief-Desire-Intention (BDI) structure is explicitly proposed, which can perform behavioral preferences learning activities in the LC process. In addition, driving behavioral preferences learning strategies are introduced which can gradually grasp essentials in driver’s subjective judgments in decision-making of the LC process and make the LC process more safety and scientific. Specifically, a conceptual model of the agent, driving behavioral preferences learning-BDI (DpL-BDI) agent is introduced, along with corresponding functional modules to grasp driving behavioral preferences. Furthermore, colored Petri nets are used to realize the components and scheduler of the DpL-BDI agents. In the end, to compare with the traditional LC parameters’ learning methods (such as: the least squares methods and Genetic Algorithms), a kind of LC problems is suggested to case studies, testing and verifying the validity of the contribution.