CCE&D: A Configuration Failure Prevention Method for Autonomous Driving Systems

The correct setting of software configuration items is essential for improving software stability and ensuring safe, reliable operation. By contrast, potential configuration errors can have serious negative effects on software operation and even cause catastrophic consequences. Compared to traditional software, autonomous driving systems involve large amounts of data acquisition, processing, and real-time decision-making, and thus have a higher degree of configurability, making them more susceptible to safety issues from configuration errors. Most previous work on configuration failure diagnosis for autonomous driving systems focused on passive diagnosis after failure occurrence, making it difficult to detect potential untriggered configuration failures during system operation. In this paper, we propose CCE&D, which automatically infers configuration constraints from source code, detect configuration failures prior to configuration-specific deployment, preventing their occurrence in autopilot systems. Experimental results show the constraint rules covers 75% of the platform’s total configuration item constraints with 98.9% accuracy. Meanwhile, the accuracy of configuration error detection reaches 96.39%, and the purpose of configuration fault prevention is achieved.