Cooperative Intelligent Transport System (C-ITS) applications assume the availability of a reliable and accurate positioning system. Even if suitable to most Day-1 applications (e.g. route navigation), the Global Navigation Satellite System (GNSS) accuracy, reliability and availability are clearly not sufficient for more demanding Day-2 applications (e.g., highly autonomous driving, advanced safety services including vulnerable road users warning, etc.), which would require a consistent sub-meter localization accuracy regardless of operating conditions. To bridge this gap, Cooperative Localization (CLoc) has been recently identified as a promising strategy. Accordingly, mobile nodes can help each other by exchanging location data (typically, their own position estimates or raw GNSS data), acquiring range-dependent metrics over their respective radio links and finally, fusing the various sources of information. However, conventional CLoc solutions may be partly unsuitable within the context of vehicular ad hoc networks (VANETs), which comes along with unprecedented challenges such as specific mobility patterns, practical operating trade-offs with complexity and vehicle-to-vehicle (V2V) communication capabilities, or even fusion optimality when multiple measurement modalities are available at the vehicles. Thus, one central related research question is as follows: Can the Day-2 sub-meter localization accuracy be met through CLoc strategies between connected vehicles?"
In this thesis, following a gradually complex approach, we aim at evaluating how and in which conditions position information from neighboring vehicles and/or associated V2V measurements may improve localization accuracy and resilience. We first develop a generic fusion-based CLoc framework, which can rely on various vehicle-to-everything (V2X) and embedded sensor technologies. We then apply this framework to the standard ITS-G5 Cooperative Awareness Messages (CAMs), and show that it is possible to benefit from neighboring position information and from received signal strength-based range estimates to enhance local accuracy. On this occasion, we also make concrete proposals to handle messages/data synchronism (through mobility-based predictions), as well as to reduce both complexity and V2V communication footprint (through links/neighbors selection, messages approximation and transmission control). Next, we extend this framework to integrate more accurate V2V measurements based on the impulse radio ultra-wide bandwidth (IR-UWB) technology, while dealing with fusion filter overconfidence and error propagation issues. Finally, under even more challenging conditions with GNSS depraved neighbors or in tunnel conditions, we considered the assistance of extra onboard sensors (inertial unit, wheel speed sensor, camera-based lane detector, etc.), as well as static roadside units (RSUs). The proposed framework and methodology show to typically improve the localization accuracy from 2 m to below 30 cm in 80% of the cases. The proposed framework has been tested analytically and through simplified simulations first, then on realistic mobility data, and finally on real data from a small-scale field test.
