Lyu, Hao (Southeast University), Liu, Wan (Research Institute for Road Safety of MPS), Wang, Ting (Tongji University), Yue, Quansheng (Southeast University), LI, JUNYAO (Southeast university), Guo, Yanyong (Southeast University)

PreSimNet: A Unified Framework for Predicting and Simulating Car-Following Behaviors in Mixed Traffic Flow

Scheduled for presentation during the Invited Session "S19b-Artificial Transportation Systems and Simulation" (TH-EA-T19), Thursday, November 20, 2025, 13:50−14:10, Surfers Paradise 1

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

The coexistence of autonomous vehicles (AVs) and human-driven vehicles (HVs) in mixed traffic flow introduces unprecedented complexity to vehicular interaction dynamics, particularly in multi-type car-following scenarios. However, existing research often focuses on limited interaction subsets, without reflecting a comprehensive understanding of mixed traffic behavior. More importantly, both trajectory prediction and car-following simulation rely on trajectory encoding information, and their potential for integration has not been fully explored. To address this, we propose PreSimNet, a physics-informed deep learning framework that seamlessly integrates trajectory prediction and car-following simulation. Firstly, we design a type-guided trajectory feature learning module in PreSimNet, utilizing a large-scale dataset of over 900 hours of real-world driving data, including four interaction types (AV-HV, AV-AV, HV-AV, HV-HV), to capture interaction subtleties. On this basis, we incorporate kinematic priors and cross- attention mechanisms to construct an accurate trajectory prediction module. In a synchronized and integrated manner, a domain-knowledge-driven mixture-of-experts approach is used to develop an adaptable and informed car following simulation module. Experimental results demonstrate that PreSimNet significantly outperforms state-of-the-art baselines in both trajectory prediction and simulation tasks. This result further confirms the complementary advantages demonstrated by the unified PreSimNet framework. This work provides a powerful toolkit for digital twin modeling in mixed traffic flow applications.