| Abstract: |
This report proposes a novel intelligent test case generation framework designed to enhance the safety of Autonomous Driving Systems (ADS) by addressing current limitations in scenario diversity, environmental adaptability, and standardization. The methodology unfolds in three key stages. First, to improve scenario coverage, we reconstruct critical “corner cases” from real-world accident reports. By utilizing panoptic segmentation and cross-referencing accident data, we rebuild these high-risk scenarios in simulators (e.g., Apollo/Carla) and apply parameter mutation to generate extensive test sets. Second, to address data scarcity in complex urban environments, we leverage open-source datasets and traffic flow prediction to simulate region-specific urban traffic. This allows for low-cost, high-fidelity testing of ADS adaptability under dynamic conditions. Third, we implement the ASAM OpenX standards (OpenDRIVE, OpenSCENARIO, OpenCRG) to ensure industry-wide compatibility. We propose fine-tuning Code Large Language Models (Code LLMs) to automate the generation of standard-compliant XML files, thereby enabling the efficient creation of complex, highly dynamic interaction scenarios. |
下载中心
会议室预定
电子邮箱