Use of Particle Filters for Cooperative GNSS Positioning

In the context of vehicle-positioning applications, Global Navigation Satellite Systems (GNSSs) have remarkable role. However, these applications have very strict safety requirements, thus needing an improvement in the performance of such positioning systems. This necessity has led to the development of Cooperative Positioning methods, also thanks to the recent rise of Vehicle-To-Vehicle (V2V) communication. The aim of such methods is in fact to improve both the accuracy and precision of the stand-alone positioning system by exploiting the exchange of relative ranging information by a network of vehicles. Moreover, in spite of the fact that the positioning problem is non-linear (i.e. trilateration), many solutions approach it by means of linearization (i.e. Least Mean Square, Extended Kalman Filter). These methods may also introduce errors by assuming that general probability distributions of the input measurements are Gaussian distributions. However, the error of these relative distances derived from the measurement exchanged by the vehicles have shown to be, in general, not Gaussian distributed, leading to a reduction in performance due to the mismodelling of such probability distributions. This has prompted the study of other solutions (i.e. Particle Filter), able to handle the problem without linearization, and also the non-Gaussian distribution of the measurement errors. Most importantly, these distributions are also non-stationary, thus requiring a real-time estimation in order to allow the Particle Filter to provide the best possible solution. In particular, the errors can be affected by a bias, and it is the aim of the thesis to understand whether the relative motion of the vehicles can affect the bias introduced in the measurements they exchange. A further goal is to design an adaptive algorithm to select optimal likelihood functions, based on the relative position, motion and GNSS measurements of the vehicles, in order to improve the estimate provided by the Particle Filter. Finally, the performance of the previously mentioned solutions is evaluated, to assess if the use of the Particle Filter is justified even when the cooperative measurements can be approximated to be Gaussian distributed. Eventually, an optimized integration of cooperative ranging measurements is performed in order to complement satellite-based measurements. This approach aims at compensating for the limited availability and high geometrical dilution of precision that are frequently experienced in urban environments. With this intent, an Agent Network (AN) is implemented, and scenarios in which such agents are static or moving are both studied, in order to gain a better understanding of the effect that their relative motion can cause on the distribution of the measured distances. The collaboration between agents is implemented with an exchange, at each time instant, of both their estimated positions and their available GNSS measurements (i.e. Doppler and pseudoranges). Thanks to these, inter-agent distances are then computed by means of Weighted Least Square Double Difference method. The integration of this additional information is expected to provide an improvement in terms of accuracy of the positioning solution.