Simulating Realistic Pedestrian Behaviors in the Context of Autonomous Vehicles in Shared Spaces

摘要

In this PhD thesis, we investigate how pedestrians will behave in a shared space with an autonomous vehicle, by modeling and simulating realistic pedestrians' behavior. Our approach integrates agent-based social simulation with standard robotic tools. We proposed an agent-based model through extensions of the Social Force Model. We implemented this model as an open source tool to simulate the interactions between pedestrians and an autonomous vehicle in shared spaces. Simulation results, when compared with available data, show that combining an agent-based model and robotics is promising for handling real-world scenarios. The proposed model and simulator are used: 1) to simulate various shared space scenarios with heterogeneous and realistic pedestrians' behaviors; 2) to provide a simulated environment to test autonomous vehicle navigation strategies; 3) to reproduce real-world scenes and predict pedestrians trajectories around the autonomous car in real time.