Simulation has been considered as one of the key enablers on the development and testing for autonomous driving systems as in-vehicle and field testing can be very time-consuming, costly and often impossible due to safety concerns. Accurately modeling traffic, therefore, is critically important for autonomous driving simulation on threat assessment, trajectory planning, etc. Traditionally when modeling traffic, the motion of traffic vehicles is often considered to be deterministic and modeled based on its governing physics. However, the sensed or perceived motion of traffic vehicles can be full of errors or inaccuracy due to the inaccurate and/or incomplete sensing information. In addition, it is naturally true that any future trajectories are unknown. This paper proposes a novel modeling method on traffic considering its motion uncertainties, based on Gaussian process (GP). A probability distribution function is employed to represent traffic vehicles’ future trajectories, which are further classified based on Gaussian Mixture Model (GMM) into typical motion trajectories. Then the GP-based motion model is built from the typical motion trajectories. With this model, any potential trajectories of traffic vehicles can be simulated by sampling the GP conditional distribution. The experiment has been performed in a high-fidelity driving simulator with a full-motion base. The results have demonstrated that the proposed GP-based model can faithfully represent the uncertainties of traffic vehicles motion, thus, is suitable for the high-fidelity simulation of autonomous driving systems.
Due to current capacity constraints, printed versions of our publications - including standards, technical papers, EDGE Reports, scholarly journal articles, books, and paint chips - may experience shipping delays of up to four to six weeks. We apologize for any inconvenience.