Norm-based distributed controllers in a Multi-Agent System

Abstract

In the near future, autonomous vehicle will take part of daily traffic. These autonomous vehicles have private goals they want to achieve, which might conflict with the goals of the traffic system designers, such as safe roads or a high throughput. To ensure that goals of the system are met, the vehicle behavior needs to be influenced. Since the vehicles are autonomous, their behavior can not be controlled directly. One way of influencing behavior indirectly is through norms. A norm describes desired behavior for a vehicle. Vehicles who do not comply with the norm receive a sanction, such as a monteray fine. The norms are issued to the vehicles by traffic control systems, which monitor the vehicles and sanction them when norm violations are detected.

In this thesis, we present a formal model and a corresponding implementation for a norm-based multi-agent system for traffic. We created a formal model of vehicles, traffic regulations and traffic controllers and implemented this in an extension of a state of the art traffic simulator named SUMO. We also performed experiments to demonstrate how our system performs in various traffic scenarios.