A formal testing method for multi-agent systems using colored Petri nets

Abstract

Autonomy in software, a system’s ability to make decisions and take actions independently without human intervention, is a fundamental characteristic of multi-agent systems. Testing, a crucial phase of software validation, is particularly challenging in multi-agent systems due to its complexity, as the interaction between autonomous agents can result in emergent behaviors and collective intelligence, leading to system properties not found in individual agents. A multi-agent system operates on at least three main dimensions: the individual level, the social level, and the communication interfaces. An organizational model formally defines a multi-agent system’s structure, roles, relationships, and interactions. It represents the social layer, capturing agents’ collective dynamics and dependencies, facilitating coherent and efficient collaboration to achieve individual and collective goals. During the literature review, a gap was identified when testing the social layer of multi-agent systems. This paper presents a testing approach by formally introducing steps to map an organizational model, here \(\mathcal {M}\)oise\(^+\), into a colored Petri net. This mapping aims to generate a formal system model, which is used to generate and count test cases based on a coverage criterion. Finally, a use case called Inspector was presented to demonstrate the method by generating test cases, executing the test, and identifying execution errors.