Intelligent lane-changing and smart evasion strategy based on safety of the intended functionality

To address the safety risks caused by environmental uncertainty during lane-changing of autonomous vehicles in complex dynamic environments, this paper proposes a DCU-RRT intelligent lane-changing strategy based on an improved RRT. The strategy introduces von Mises distribution to guide directional sampling, combined with safety region constraints and vehicle-speed-adaptive step size to improve search efficiency and path quality. Furthermore, a safety evaluation model based on spatiotemporal envelope overlap is constructed to perform multi-attribute decision-making on candidate paths from three dimensions: spatiotemporal conflict detection, comfort, and timeliness. Subsequently, a constrained least squares method is employed to fit discrete path points into quintic polynomial curves, ensuring path continuity and dynamic feasibility. To enhance environmental adaptability, the system utilizes Kalman filtering to predict the motion states of surrounding vehicles in real time and designs a prediction-point-update-point alternation mechanism, which actively replans when the spatiotemporal evaluation value of the path decreases and possesses the capability to switch to an evasive path in real time. Hardware-in-the-loop experimental results demonstrate that: compared with traditional RRT, DCU-RRT reduces average search time by 33.1% and the number of iterations by 26.6%; under non-emergency conditions, the lane-changing success rate of this strategy improves by approximately 10.6% compared with the traditional quintic polynomial method; under emergency conditions, the lane-changing success rate improves by up to 28.1% compared with the traditional quintic polynomial method and by approximately 12.6% compared with dynamic RRT, achieving a maximum safety rate of 95%; in highway emergency scenarios, the lane-changing success rate of this strategy reaches 94.5%, significantly outperforming Model Predictive Control at 75.4%. The proposed strategy effectively addresses the SOTIF issues of autonomous vehicle lane-changing in complex dynamic environments.