Abstract

Autonomous vehicles can reduce or solve problems such as traffic accidents, congestion, and CO2 emissions. Although there is a lot of interest in autonomous vehicles, practical applications are still rare, mainly because of the limited sensor technology. Expectations are that the driving support systems that are currently introduced in the passenger-car market will gradually evolve and take over more and more driving tasks in time. The ultimate driving support system will be a fully autonomous and intelligent vehicle that automates all driving tasks.

In this thesis, we present a general model of an intelligent agent that can control an autonomous vehicle in different environments. An intelligent agent is an autonomous, computerised entity capable of sensing its environment and acting intelligently based on its perception. Our driving agent performs tactical-level driving which consists of all driving manoeuvres that are selected to achieve short-term objectives.
Besides controlling real-life vehicles, intelligent agents can also be used to control simulated vehicles in traffic simulations. We have expanded the functionality of our driving agent so that it can simulate different driving styles. To ensure fast reaction times, the agent’s driving task is divided in several competing and reactive behaviour rules. Each set of behaviour rules deals with a different aspect of driving in an urban environment. The agent is implemented and tested in a prototype traffic simulator program. The simulator models multi-lane roads, intersections, traffic lights, light controllers and vehicles. Every vehicle in the simulator is controlled by a driving agent and each agent is equipped with its own set of behavioural parameters to simulate individual drivers. Preliminary experiments have shown that the agents exhibit human-like behaviour ranging from slow and careful to fast and aggressive driving.