Implementation and Simulation of Sequential Adverse Condition Scenarios for Autonomous Driving

Abstract

Establishing an environment that allows for the quantitative evaluation of the ability of autonomous driving systems to respond to real-world adverse conditions is crucial to ensuring their safety and reliability. This study proposes a dynamic scenario-based simulation framework that simulates complex and sequential hazardous scenarios frequently encountered in actual road environments. The proposed scenarios are implemented based on real-world locations, including the Gwangan Bridge and Sinsundae Underpass in Busan, Republic of Korea, and the Autonomous Vehicle Test Road at Korea Intelligent Automotive Parts Promotion Institute (KIAPI) in Daegu. The proposed framework encompasses various adverse conditions, such as partial or complete loss of global navigation satellite systems (GNSS) signals in underpasses and tunnels, degraded camera and light detection and ranging (LiDAR) sensor performance due to heavy rainfall and dense fog, and blind spot formation caused by surrounding vehicles. A notable feature of the proposed framework is its ability to realize continuous and realistic transitions between different conditions. For example, entering a tunnel and experiencing GNSS signal loss, immediately followed by exposure to heavy rainfall upon exiting the tunnel during regular road driving. The simulated scenarios enable the evaluation of how autonomous driving systems respond to and manage risks in real-world environments.