Wen, Xiao (Hong Kong University of Science and Technology), Jian, Sisi (The Hong Kong University of Science and Technology)

Multi-Task Deep Reinforcement Learning for Socially Aware Car-Following Control in Mixed Traffic

Scheduled for presentation during the Regular Session "S38a-Towards Scalable and Trustworthy AI in Connected Mobility" (FR-LM-T38), Friday, November 21, 2025, 10:50−11:10, Coolangata 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Current autonomous vehicle (AV) car-following control research prioritizes AV performance in isolation, neglecting the interactions with the following human-driven vehicles (HDVs). This omission can result in socially non- compliant AV behaviors, increasing unpredictability for the following HDVs and elevating rear-end collision risks. We propose a socially compliant car-following controller for AVs to address this gap based on the well-known social value orientation (SVO). By exploiting the continuous spectrum of SVO, we develop a multi-task deep reinforcement learning (MTDRL) model consisting of: (1) a single-task DRL backbone, and (2) a curriculum learning-based multi-task scheme for efficient knowledge transfer across SVO variations. Experiments demonstrate that MTDRL improves oscillation dampening, driving comfort, and fuel efficiency compared to single-task DRL. In addition, the model can be generalized well to unseen SVO values, highlighting its adaptability in mixed traffic environments