Data-Driven Identification and Analysis of Road Test Scenarios for Self-Driving Minibuses

Qianli, Ma; Tong, Wang; Jinglei, Chen; Hongyu, Chen

doi:2026.27.3.987

International Journal of Automotive Technology > Volume 27(3); 2026 > Article

Connected Automated Vehicles and ITS, Vehicle Dynamics and Control

International Journal of Automotive Technology 2026;27(3): 987-1006.
doi: https://doi.org/10.1007/s12239-025-00345-2

Data-Driven Identification and Analysis of Road Test Scenarios for Self-Driving Minibuses

Ma Qianli^1,2, Wang Tong^1,2, Chen Jinglei^1,2, Chen Hongyu³

¹Collaborative Innovation Center for Transport Studies, Dalian Maritime University, Dalian, 116026, Liaoning, China
²School of Maritime Economics and Management, Dalian Maritime University, Dalian, 116026, Liaoning, China
³Dalian Huarui Heavy Industry Group Co. Ltd., No. 169, Ba Yi Road, Xigang District, Dalian, 116016, Liaoning, China

Corresponding Author. Ma Qianli , Email. qianlima@dlmu.edu.cn

Received: April 1, 2025; Revised: June 5, 2025 Accepted: July 29, 2025. Published online: September 3, 2025.

ABSTRACT

Autonomous minibuses are expected to be the first high-level autonomous vehicles used commercially. Before deployment, thorough testing and evaluation are crucial for ensuring safety and reliability. This paper proposes a data-driven framework for identifying and analyzing road test scenarios of autonomous minibuses, thereby improving the efficiency and coverage of testing. The framework consists of three phases. Initially, driving data are split into scenario samples through sliding window processing. Next, an LSTM-based method extracts features from multivariate time series, including speed, acceleration, and steering wheel angle, facilitating the recognition of critical scenarios. Finally, K-Means clustering and similarity analysis with human-annotated scenarios are utilized for automatic scenario categorization. Experiments demonstrate that the framework can effectively identify representative driving scenarios through temporal pattern recognition, with clustering results closely matching those of expert-annotated scenarios. Compared to traditional manual annotation, this approach achieves significant efficiency gains while fully covering key traffic situations. In conclusion, this data-driven approach provides a systematic pipeline for generating automated driving scenarios, transforming raw sensor data into structured scenario libraries. It provides an efficient solution for accelerated testing of autonomous vehicles.

Key Words: Urban transportation · Data-driven · Autonomous minibus · Scenario recognition · Road test