Jiang, Yanbo (tsinghua university), Wang, Yuning (Tsinghua University), Wang, Jiahao (Tsinghua University), Ke, ZeHong (Tsinghua University), Qingwen, Meng (Tsinghua university), Xu, Qing (Tsinghua University), Wang, Jianqiang (Tsinghua University)

A Personalized Interactive Autonomous Vehicle Decision-Making Method Considering Driving Style Recognition

Scheduled for presentation during the Invited Session "Towards Human-Inspired Interactive Autonomous Driving II" (ThBT2), Thursday, September 26, 2024, 15:10−15:30, 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

Decision-making is one of the critical modules of autonomous vehicles, and how respond to other agents under interactive scenarios is still a challenge. Although current decision-making methods have managed to conduct longitudinal and lateral plannings, the adaption to driver feature diversity is still not sufficient, failing to consider individual driving demands. To enhance adaptability of autonomous driving decision-making, this paper presents a personalized interactive autonomous vehicle decision-making method that integrates driving style recognition. By analyzing driver history driving trajectories from open datasets, related driving factors are extracted. To reduce the problem dimension, Principal Component Analysis and Factor Analysis are applied so that key parameters can be filtered. With the input of extracted key factors, a K-means clustering module is proposed to recognize the driving style which is classified into conservative, general, and aggressive. After identifying driver style, we design a decision-making fusion strategy which merges individual driver identification into both lateral and longitudinal control methods based on MPC and polyline planning. We validated the effectiveness of the method through real-world vehicle testing, and the results demonstrate that our method can accurately recognize driving styles, thereby achieving highly human-like autonomous driving decisions.