Decentralized Runtime Norm Enforcement

Abstract

Smart roads are an electronically enriched form of highways that aim to maximize the throughput and safety of traffic. If a vehicle is not complying with its instructions, then an appropriate reaction such as a fine is required. This gives us two basic tasks for a smart road. On the one hand does it have to monitor the traffic in order to determine the instructions for vehicles and whether they comply with their instructions, on the other hand does it have to process the violations of instructions. In this thesis we view the tasks of a smart road as an example of the more generic task of making sure that agents behave according to preset guidelines. A specification of how agents ought to behave and the measures when they fail to comply, is what we call a norm. In our traffic example we may consider a norm to be comparable with a traffic rule. The task of a smart road can be reformulated as `the enforcement of one or more norms'. The traffic situation on one road may affect the situation on another. Therefore, it is often appropriate to consider norms not in the context of a single smart road but in a network of them. Smart roads have to collaborate in such cases in order to enforce the norms. We call this decentralized enforcement. We are particularly interested in the limitations and possibilities of decentralized norm enforcement when we want to apply it in an application. To this end, we focus specifically on decentralized runtime norm enforcement. Our main conclusions and contributions are as follows. Decentralized runtime norm enforcement is composed of monitoring and control. Monitoring for decentralized runtime norm enforcement can be equated to verifying whether a system satisfies a linear temporal logic (LTL) formula. Several proposals exist for runtime LTL verification, to which we have added two complementary proposals. We also discussed security and robustness for a decentralized monitor. The control task of decentralized runtime norm enforcement is comparable to the control of discrete event systems. We took a runtime controller model from the literature of discrete event control systems and have shown how this model can be reapplied for norm enforcement. A typical property of decentralized enforcement is that multiple controllers can operate in a concurrent manner, such as multiple smart roads that are coordinating traffic concurrently. We describe how the model for controllers can be expanded to the concurrent application of controllers, which results in a collaborative controller. It is important to consider conventional programming paradigms in order to promote the development of agent systems with norm enforcement. We proposed design patterns for object-oriented programming which capture often reoccurring solutions in the agent programming literature. We also showed how an object-oriented implementation can be expanded with norm enforcement using aspect-oriented programming. Finally, we made an example simulation of a smart roads scenario. This simulation illustrates how we may view the task of a smart road as the decentralized runtime enforcement of norms.