MARL-Based High-Risk Multivehicle Scenario Generation for Autonomous Vehicle Safety Testing
- 15 August 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (22) , 46928-46940
Abstract
Advancements in autonomous driving technology have been substantial, yet ensuring the safety and reliability of autonomous vehicles (AVs) in high-risk situations remains a critical challenge. Traditional mileage-based testing is often inadequate for capturing rare but safety-critical events, leading to a growing emphasis on scenario-based virtual simulation. This study proposes a novel method for generating high-risk multivehicle scenarios using multiagent reinforcement learning (MARL) and in-depth crash data. A total of 567 real-world crash scenarios were extracted from the China in-depth mobility safety study-traffic accident (CIMSS-TA) database and reconstructed to include both crash-involved and noncrash-involved vehicles that may have influenced the crash. A multiagent deep deterministic policy gradient (MADDPG) algorithm is employed to model interactions among these vehicles and the AV under test, with custom-designed reward functions guiding adversarial agent behavior. The generated scenarios were validated using surrogate safety measures, assessing both risk levels and crash severity. Experimental results demonstrate that the proposed method produces significantly more complex and risk-intensive scenarios than baseline approaches, thereby exposing potential weaknesses in AV decision-making. Notably, the inclusion of noncrash-involved vehicles substantially elevates scenario difficulty, highlighting their critical role in realistic AV safety assessment. This framework offers a scalable approach to augmenting scenario diversity and enhancing the credibility of AV testing protocols.Keywords
Funding Information
- National Key Research and Development Program of China (2023YFB2504700)
- science and technology innovation Program of Hunan Province (2023RC3143)
- China Postdoctoral Science Foundation (2025M771647)
- Postdoctoral Fellowship Program of CPSF (GZB20250421)
- Natural Science Foundation of Hunan Province (2024JJ5030)
- National Natural Science Foundation of China (62473381, 52502435)
This publication has 44 references indexed in Scilit:
- Pre-crash scenarios for safety testing of autonomous vehicles: A clustering method for in-depth crash dataAccident Analysis & Prevention, 2024
- How would autonomous vehicles behave in real-world crash scenarios?Accident Analysis & Prevention, 2024
- Identifying typical pre-crash scenarios based on in-depth crash data with deep embedded clustering for autonomous vehicle safety testingAccident Analysis & Prevention, 2023
- Enhanced index of risk assessment of lane change on expressway weaving segments: A case study of an expressway in ChinaAccident Analysis & Prevention, 2022
- Multi-vehicle interaction safety of connected automated vehicles in merging area: A real-time risk assessment approachAccident Analysis & Prevention, 2022
- Waymo simulated driving behavior in reconstructed fatal crashes within an autonomous vehicle operating domainAccident Analysis & Prevention, 2021
- A novel approach to set driving simulator experiments based on traffic crash dataAccident Analysis & Prevention, 2021
- Mining and comparative analysis of typical pre-crash scenarios from IGLADAccident Analysis & Prevention, 2020
- A clustering approach to developing car-to-two-wheeler test scenarios for the assessment of Automated Emergency Braking in China using in-depth Chinese crash dataAccident Analysis & Prevention, 2019
- In search of the severity dimension of traffic events: Extended Delta-V as a traffic conflict indicatorAccident Analysis & Prevention, 2017