An agent-based traffic simulation framework to model intelligent virtual driver behaviour

This paper presents an agent-based traffic simulation framework that supports intelligent virtual driver behaviour. The framework exploits concepts used in Artificial Life (ALife), Artificial Intelligence (AI) and Agent technology to model the inherent unpredictability and autonomous behaviour of drivers within traffic simulation models. Each driver agent in our system contains knowledge and a decision-making mechanism, both of which are based on heuristics. This approach replaces some of the prescriptive nature of driving simulation models by allowing behaviours to emerge as a result of individual driver agent interactions. The framework also contributes to accident analysis by improving current limitations in which accident investigation methods concentrate on the events themselves, rather than pre-crash influences. Within this context, the framework provides an opportunity to increase the understanding of accident causation factors, to examine alternative traffic scenarios (what if analyses) and methodology to obtain quantitative estimates of accident risk. Current implementation results show that driver agents within the integrated simulation are able to perceive other drivers’ speeds and distances, avoid collisions, perform realistic vehicle following, and demonstrate emergent traffic flow. A major application area for this framework includes the evaluation of vehicle, highway and road user factors that precede a collision, or near misses.