SimLKAS: a simulation-based framework for the verification and validation of lane keeping assistance systems

Lane Keeping Assistance Systems (LKAS) significantly enhance vehicle safety by maintaining lane centering through automated steering control. Evaluating LKAS performance in real-world conditions is challenging, as rare and extreme scenarios are difficult to capture systematically. Moreover, real-world testing is often time-consuming, costly, and lacks precise control over critical environmental factors, limiting its effectiveness for comprehensive assessment. However, existing simulation-based evaluations, such as those in CARLA, assume perfect pavement marking detection and neglect lateral deviations, limiting real-world applicability. To address these challenges, this study introduces SimLKAS, a simulation-based framework implemented in CARLA for systematic verification and validation of LKAS under diverse and challenging environments. The framework comprises three modules: context-based pavement marking perception, feedback-based loop controller, and ensemble learning-based performance evaluation. A key advantage of this modular framework is its analytical traceability for assessing the propagation of data inaccuracies and modeling errors across layers, along with its algorithmic flexibility for integrating alternative algorithms at each stage. Extensive experiments confirm that SimLKAS generates diverse testing conditions while reproducing realistic lane-keeping patterns consistent with previous studies, validating its effectiveness for LKAS assessment. The results, analyzed using Random Forest regression and SHAP analysis, indicate that heavy rain, high speed, and curved roads significantly contribute to increased lateral deviation in daytime conditions, whereas nighttime performance is strongly influenced by streetlights and vehicle headlight settings. This framework provides researchers and practitioners with a scalable, repeatable, and controlled platform for benchmarking LKAS algorithms, optimizing performance, and enhancing safety validation, thereby supporting advancements in autonomous driving technology.