A Fallback Localization Algorithm for Automated Vehicles Based on Object Detection and Tracking

Integrating Automated Vehicles (AVs) into everyday traffic is an ongoing challenge. Ensuring the safety of all involved agents, even in the presence of system failures, is crucial, especially in urban environments. This paper introduces a fallback-oriented localization algorithm for AVs designed to operate during main localization source failures. The method leverages stationary vehicles as dynamic landmarks, identified through the perception module, despite their initially unknown positions. By tracking relative positions before failure and applying trilateration, the algorithm estimates the ego vehicle's position. The proposed algorithm is evaluated through simulations, a real-world dataset, and practical tests on two vehicle models. The results include an average trajectory error of 0.62 m and 1.58 deg compared to the ground truth over different fallback maneuvers. This translates into an average relative translational error of 1.65 % and a relative rotational error of 0.05 deg/m, improving the performance of an IMU-based dead reckoning and, hence, providing localization for performing safe stop maneuvers.