Download files
Abstract
Driving is naturally an interactive task where individual drivers are continuously manoeuvring based on expectations and beliefs regarding actions of other drivers. However mathematical models of interaction are seldom used for modelling driver manoeuvres. The dominant modelling methods rather assume there is a one-dimensional interaction between the stimuli provided by a driver, and the response of any conflicting drivers. This oversight of the inter-relation between conflicting drivers’ actions fails to scientifically evaluate key aspects of the safety and efficiency of driver interactions: behavioural norms of interaction, and moral hazards of interaction.
Improving modelling of driver conflicts is important to evaluate the safety of existing road infrastructure and proposed technology measures that aim to improve driver safety. Thus Game Theory which is a framework for mathematical models of interaction has been emerging as a research area within traffic analysis. Given this, game theoretic literature for driver manoeuvres still poses a lacuna in the investigation of modelling sub-processes that are critical for accurate game theoretic predictions against reality. These modelling aspects are the specification of payoff functions for decisions, and methods to mathematically calculate interaction game solutions.
The research project within this thesis accordingly develops game theoretic models that investigate the importance of risk attitude and risk perception parameters in payoff functions for interactive driver manoeuvres, and efficacy of a Quantal Response Equilibrium game solution.
An empirical approach is used to calibrate and verify the significance of the proposed game theoretic models and frameworks against observed driver interactions. In particular, field data of an experiment conducted in an experimental economics laboratory is used, as well as GPS trajectory of vehicles merging at an on-ramp along Interstate 80 in Emeryville, California. The studies find that the introduction of risk-related parameters in payoff functions for interactive decisions is important to explain observed driver interactions and a Quantal Response Equilibrium game solution is able to provide good fit to observations.
The key contributions of the research project are novel approaches for improved modelling of interaction in driver manoeuvres. The modelling approaches help to more accurately evaluate existing and proposed measures towards traffic safety and efficiency.