Motion planning based on learning models of pedestrian and driver behaviors

Autonomous driving has shown the capability of providing driver convenience and enhancing safety. While introducing autonomous driving into our current traffic system, one significant issue is to make the autonomous vehicle be able to react in the same way as real human drivers. In order to ensure that an autonomous vehicle of the future will perform like human drivers, this paper proposes a vehicle motion planning model, which can represent how drivers control vehicles based on the assessment of traffic environments in the real signalized intersection. The proposed motion planning model comprises functions of pedestrian intention detection, gap detection and vehicle dynamic control. The three functions are constructed based on the analysis of actual data collected from real traffic environments. Finally, this paper demonstrates the performance of the proposed method by comparing the behaviors of our model with the behaviors of real pedestrians and human drivers. The experimental results show that our proposed model can achieve 85% recognition rate for the pedestrian crossing intention. Moreover, the vehicle controlled by the proposed motion planning model and the actual human-driven vehicle are highly similar with respect to the gap acceptance in intersections.