Building a Driver Model Using Risk Potential Theory in Collision Avoidance Situation

Kikuta, R., Carruth, D. W., & Kageyama, I. (2023). Building a Driver Model Using Risk Potential Theory in Collision Avoidance Situation. Institute of Industrial and Systems Engineers (IISE) Annual Conference 2023. New Orleans, LA: Institute of Industrial and Systems Engineers (IISE).

This research describes the construction and evaluation of a driver model for vehicle longitudinal control that can avoid collisions with pedestrians. While the priority for the driver model is to avoid collisions, this work also considers the potential trade-off between collision avoidance, comfort for vehicle occupants, and the effects on traffic flow. Three driver models were developed based on risk potential (RP) methods: a conservative RP model, a relative velocity RP model, and a Pedestrian Potential Position (PPP) RP model. The PPP model considers pedestrian dynamics such as position, speed, and direction of movement in its assessment of RP. The three driver models were implemented in MATLAB/Simulink and performance of the models was evaluated in simulations of nine situations involving vehicle-pedestrian interaction. The models were compared pairwise according to risk of injury to the pedestrian, comfort for vehicle occupants, and false alarms. The PPP model outperformed the other RP models in situations that have a severe risk of collision. In safe situations, the velocity model outperformed the other RP models. The PPP model was potentially over-cautious when approaching pedestrians standing near the side of the road without the intention to cross. Overall, the proposed models are expected to improve the performance of RP driver model reducing the risk of injury, occupant discomfort, and the number of false alarms.