Optimal Collaborative Positioning

Dr.-Ing. Nicolas Garcia Fernandez

Main-Supervisor: S. Schön; Co-Supervisor: C. Heipke


Applications such as Intelligent Transportation Systems (ITS) heavily rely on the continuous availability of Position, Navigation and Timing (PNT) information in order to guarantee the safety of the users. This goal depends largely on the availability of Global Navigation Satellite Systems (GNSS) signals. However, it is well known that although GNSS represents a fairly continuous and time stable source of information for the localization, its performance is degraded in urban areas. To overcome the weakness of the GNSS standalone solution, the GNSS measurements are fused with measurements carried out with additional sensors, that increase the awareness of the ego vehicle with respect to the elements of the environment. The development of Vehicle to Everything (V2X) communication protocols enable the expansion of the principle of multi-sensor estimation to multi-vehicle approaches, leading to Collaborative Positioning (CP). CP is a network-based navigation technique, that is suitable to increase the performance of the navigation solution, and the robustness against outliers and sensor failures. The concept of CP resembles a geodetic network in which the nodes (vehicles) are moving. Thus, as in the case of geodetic static networks, the purpose of this project is to aid the understanding of the dynamic network sensitivity with respect to changes in its geometry, topology or changes in the motion of the object. Later, the findings can be used in order to determine the steps to follow in order to achieve an optimum solution. For this purpose, a simulation framework is designed, suitable to cost-effectively and safely reproduce a wide range of collaborative scenarios. In this simulation framework, the measurements carried out with sensors mounted in different vehicles (GNSS receivers, IMUs, laser scanner and cameras) are fused in a Collaborative Extended Kalman Filter (C-EKF) in which the state parameters of all vehicles are simultaneously estimated. The simulation framework is used in order to investigate the following aspects: - The performance of the C-EKF with different execution strategies, namely localization, localization or SLAM. - The sensitivity of the estimation with respect to changes in the dynamic network geometry and topology. - The challenges that arise from the dynamic movement of the nodes in the network evaluation. - The evaluation of the benefits of CP with respect to single vehicle approaches. The results enable to link levels of performance achieved with the different execution techniques to specific sensor assemblies. Also, the node dynamics are confirmed as a crucial variable to be optimized in order to maximize the performance of the filter. In such time-variant systems, the characteristics of the estimation vary vastly from epoch to epoch, arising the necessity to develop process noise models, motion models and algorithms capable of adjusting to the ongoing vehicle manoeuvres. Finally, the benefits of CP with respect to single vehicle are assessed and catalogued in terms of accuracy and precision, for each individual node (local analysis) and the overall network (global analysis). The number of vehicles included in the collaborative navigation scenario is a crucial variable to take into account, being able to prove that the higher the amount of dynamic nodes (multi-sensor systems) considered, the higher the accuracy and precision of the solution.