Drivers with limited perception: model and application to traffic simulation

Abstract

We propose a model of the driver perception suited for microscopic, agent-based traffic simulations. The model includes both top-down and bottom-up perception, and takes into account the limited amount of perceptive resource which gain access to short-term memory. The driving task is split into sub-tasks, which can be activated in parallel (e.g. car following and crossroads passing). Perceived entities (percepts) as well as subtasks are ranked with respect to their subjective value, and due to the bounded perception, only the more "valuable" percepts are sent to the decision module of the cognitive model. The competition among percepts to gain access to the short-term memory simulates attentional processes. A computational implementation of the model is proposed for the driver, using agent-based modeling. It is implemented in a traffic simulation environment and allows the driver-agent to manage the conflicts and the longitudinal space in the middle of the crossroads. This way, we improve the realism of the simulation. Furthermore, this model can lead to a new way of identifying and explaining near accidents. We illustrate some benefits for a microscopic traffic simulation at crossroads in two situations. The first

AVDM: A hierarchical command-and-control system architecture for cooperative autonomous vehicles in highways scenario using microscopic simulations

“…In [105], Weyns introduced a generic perception model for situated multi-agent system where the agent's perception rely on a set of sensors, precepts (machine-readable real-object representation) and filters specific to the situation. This model inspired Ketenci to design a limited perception model for drivers in a traffic simulation [59]. Tabelpour and Mahmassani have proposed an alternative approach in [95] where sensor range and accuracy limitations were considered to create input data for the car-following model.…”

Agent Bodies: An Interface Between Agent and Environment

“…This approach enables to define a mixed bottom up and top down approach of perception [13]: the classical perception methods, which are requests from the agent to the environment, are top-down, since they are driven by the agent, may be mixed with "awareness" filters, defining percepts that are perceived by the agents event if they do not request them [25,16].…”