摘要：AbstractCooperative localization has been considered to be a suitable solution for accurate and robust localization in the field of autonomous driving. This paper considers a scenario, where a vehicle is not able to access any GNSS measurements, other than through vehicle-to-vehicle (V2V) communication. Using the V2V communication, the accurate absolute position of the neighbor vehicles can be received. Assisted with distance sensors and local inertial sensors, the local absolute position can be determined. We propose a novel estimation scheme which integrates the measurements of distance sensors, inertial sensors and neighboring vehicles’ information. A graph matching method based on the concept random sample consensus (RANSAC) is proposed to match the neighbor position with a clustered object from a LIDAR distance sensor. Sorted information is then fused into a multirate EKF to update the vehicle states, which are predicted using a strap down propagation. The proposed algorithm is evaluated in simulation and subsequent in a real-world experiment. A 2D error of less than 0.5m can be achieved in simulation as well as in the real world. The accuracy can be improved to approximately 0.2m if more neighbor vehicles are networked.
关键词：KeywordsCooperative localizationGraph matchingExtended Kalman FilterClusteringSensor fusionVehicle state estimationRANSAC
