Achieving dynamic road traffic management by distributed risk estimation in vehicular networks

Abstract

In this thesis I develop a model for a dynamic and fine-grained approach to traffic management based around the concept of a risk limit: an acceptable or allowable level of accident risk which vehicles must not exceed. Using a vehicular network to exchange risk data, vehicles calculate their current level of accident risk and determine their behaviour in a distributed fashion in order to meet this limit. I conduct experimental investigations to determine the effectiveness of this model, showing that it is possible to achieve gains in road system utility in terms of average vehicle speed and overall throughput whilst maintaining the accident rate. I also extend this model to include risk-aware link choice and social link choice, in which vehicles make routing decisions based on both their own utility and the utility of following vehicles. I develop a coupled risk estimation algorithm in which vehicles use not only their own risk calculations but also estimates received from neighbouring vehicles in order to arrive at a final risk value. I then analyse the performance of this algorithm in terms of its convergence rate and bandwidth usage and examine how to manage the particular characteristics of a vehicular ad-hoc network, such as its dynamic topology and high node mobility. I then implement a variable-rate beaconing scheme to provide a trade-off between risk estimate error and network resource usage.