Map-Aided Automotive Dead-Reckoning using Rao-Blackwellized Particle Filter

摘要

The paper presents the dead-reckoning algorithm for the vehicles. The algorithm is based on Rao- Blackwellized Particle Filter and operates in global coordinates (latitude, longitude and altitude). The key feature is the usage of buildings' footprints and road networks for position aiding. There are no restrictions on off-road driving. The algorithm fuses the data from low-cost IMU, odometer, road and buildings' maps. The system state consists of vehicle's position, its velocity and attitude, as well as the sensors' errors. The algorithm can operate in real-time on a PC with the number of particles up to 10000. This performance is achieved by utilizing the spatial trees for map data requests and error-state covariance propagation. The algorithm was tested by extensive simulations and real experiments. The average positioning error (APE) in the case of perfectly known road network and buildings' footprints did not exceed 2 meters after 10 minutes of operation in dead-reckoning mode. The maximum error did not exceed 7 meters. In the case of off-road driving and roads' and buildings' representation inaccuracies, the APE after 10 minutes of autonomous operation was ≈ 12 meters. The proposed solution allows to aid the odometry error drift and provide lane-level navigation accuracy on a long periods of operation in dead-reckoning mode.