Xing, Yining (Tsinghua University), Cai, Mengchi (Tsinghua University), Tu, Yiqian (Tsinghua University), Wang, Yuning (Tsinghua University), Jiang, Junkai (Tsinghua University), Qingwen, Meng (Tsinghua university), Wang, Jianqiang (Tsinghua University)

TREMP: Field-Based Traffic Risk Evaluation Considering Interactive Motion Prediction

Scheduled for presentation during the Invited Session "Towards Human-Inspired Interactive Autonomous Driving I" (ThAT2), Thursday, September 26, 2024, 10:50−11:10, Salon 5

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Traffic risk evaluation is a critical technology in traffic management and autonomous driving. The key challenge is to identify the potential intentions of agents that may generate risks. Traditional risk evaluation methods often rely on the current state of vehicles to infer their future intentions, which lacks accuracy. This paper proposes a traffic risk evaluation method considering interactive motion prediction (TREMP). The trajectory prediction results are used to represent the potential intentions of vehicles, and the distribution of prediction errors characterizes the spread of risks. Additionally, we utilize real-world datasets to analyze the distribution of surrounding agents and model the risks generated by vehicles on the road. By integrating the risk of each moment with the intentions of future moments, we construct a risk evaluation model considering trajectory prediction. Experiments shows improvements in both the misdetection rate of risk identification and the advance time for risk detection. Examples illustrate a strong correlation between the risks evaluation result and actual driver behavior, enabling accurate analysis of traffic risks.