Analysis of spatio-temporal trends in NZ traffic & implications for benchmarking and implementation of self-driving cars
Holiday news coverage is unfortunately never complete without reports of traffic jams and the road death toll. Autonomous vehicles are expected to solve both problems. Always sober, always alert, always obedient to the rules. In New Zealand, they have the potential to reduce the number of traffic accidents caused by drunk or distracted drivers, as well as those involving tourists who are unfamiliar with driving on the left side of the road. However, it would be unrealistic to assume that the autonomous vehicles could achieve a perfect driving record and avoid fatalities altogether. So how safe do they need to be before we can accept them onto our roads? To move this discussion from philosophy to policy we need to establish empirical benchmarks. The goal of this project is to use past road accident data to develop a simulation framework for the cost-benefit analysis of autonomous self-driving vehicles.
In the first phase of this project we will analyse the data in the New Zealand Crash Analysis System (CAS) to determine the factors most influencing the probability and severity of accident occurrence. CAS includes detailed, spatially-explicit traffic accident information which we intend to complement with other relevant information, such as weather conditions. In the second phase we intend to create a simulation framework that will empower us to formulate recommendations regarding successful deployment of autonomous vehicles. Will they be most useful in large densely populated metropolitan areas during the morning rush hour or should they better be deployed in remote areas instead? Would it make tourist trips safer? By establishing how safe human drivers are we will be able to establish safety performance standards for autonomous vehicles. Once autonomous vehicles meet this benchmark it could be argued that human driving should become illegal. Given the size and complexity of the data, the candidate is expected to utilise state-of-the art modelling, such as Bayesian spatio-temporal regression modelling, Bayesian networks, and random forests.
Supervisor: Elena Moltchanova
Key qualifications and skills
Bayesian statistics, MCMC algorithms, data mining, data science, good programming skills (R preferred)
Does the project come with funding
Final date for receiving applications
Bayesian statistics; data science