A Platform for Estimating the Relevance of Information in VANET Applications

Progress in wireless communication and sensing technologies enabled research on Vehicular Ad-Hoc Network applications that aim to disseminate useful information. Examples include safety applications such as the Emergency Electronic Brake Lights, Highway Merge Warning, or Control Loss Warning, or non-safety applications such as parking or travel time dissemination systems. A major problem in these applications is knowing when the information is relevant for a given vehicle. The knowledge of information relevance helps applications to make decisions such as when to warn the driver because of some reported safety related information or which parking location should the driver pursue given a list of available locations? The estimates of relevance can also be used to rank the information. This maximizes the use of available resources such as the communication bandwidth and allows for the most important information to be disseminated. Previously proposed methods for estimating relevance typically depend on heuristics or analytical solutions. These may be inaccurate or may not consider all the necessary factors. They are also application specific and therefore it is hard for developers to use these for novel applications. We propose a simulation based platform for developing novel VANET applications. The platform generates relevance estimator modules that can be used in deployed applications. The method for generating the modules is based on a machine learning approach. The method works by using observations of vehicles in simulations to generate training examples that are used to learn a relevance function, which estimates the relevance of the given piece of information. The dissertation presents research work on using this technique for safety and non-safety applications. It also presents an implementation of the platform for developing novel Vehicular Ad-Hoc Network safety warning applications.