Multi-Agent Multi-Sensor Collaboration for Improved Positioning in Urban Environment

In recent years, the demand for accurate and reliable positioning has increased. Especially with regard to autonomous driving, highly accurate positioning information is needed. While conventional Global Navigation Satellite System (GNSS) positioning solutions provide reliable results in open areas such as highways, this is quite different in urban environments (e.g. urban canyons, tunnels or underground car parks). For these areas, GNSS positioning must be supported by other information in order to guarantee accurate and reliable positioning. For this contribution, sensor fusion is applied to a multi-vehicle network in order to make statements about the opportunities and limitations of collaborative navigation on the accuracy and robustness of position determination. The focus is on collaboration itself and not on communication. We use a Real-Time Kinematic (RTK) algorithm supported by inter-vehicle observations from stereo imagery to aid the position estimation. For this paper, we consider real data collected in a GNSS challenging environment in Hannover, Germany. The original inter-vehicle and GNSS observations are modified into bias-free observations with Gaussian noise. As Vehicle-to-Vehicle observations, we simulate a total of 18 coordinate differences with different precisions. The estimated position results are verified with a reference solution. This approach is then used to understand the potential and limitations of collaborative positioning. We demonstrate that the improvement of the C-RTK solution as well as its magnitude are determined by the precision of the V2V observations, the amount of available GNSS observations, and the performance of the aiding agents in the network. We show that the use of a collaborative RTK leads to a max. improvement of the 3D RMSE of up to 51% compared to the single-vehicle (SV) RTK solution. The collaborative solution has a particularly positive effect on the determination of the height component. In contrast, a careful selection of collaboration partners is required, since imprecise and inaccurate aiding agents might lead to a significant deterioration compared to the SV solution.