Robust efficient ego-vehicle path prediction based on Bezier curves for autonomous driving

Abstract

Accurate ego-vehicle path prediction is essential for safety-critical functions in advanced driver assistance systems (ADAS), such as automatic emergency braking (AEB) and collision avoidance. Existing models based on Clothoid curves are typically not sufficient in expressing complex maneuvers and are not highly adaptive to various vehicle dynamics. In addition, these models struggle with accuracy in circular maneuvers and fail to use in complex paths (e.g., S-shapes). This paper proposes a novel representation of the ego-vehicle path prediction using Bezier curves. The proposed Bezier curves are composed of two Cartesian third-order polynomial functions. They are formulated efficiently to model both circular and S-shaped trajectories with high accuracy and low computational cost. Our method significantly reduces prediction error, achieving over 95% improvement in average Euclidean distance error compared to Clothoidal models along about 50 m paths in controlled circular scenarios. The proposed algorithm, designed with O(n) complexity, is suitable for real-time applications on low-power automotive hardware. Its effectiveness is demonstrated through simulation using CarMaker, and a collision estimation module for AEB is developed based on the predicted paths.