Planning while Believing to Know

Over the last few years, the concept of Artificial Intelligence (AI) has become essential in our daily life and in several working scenarios. Among the various branches of AI, automated planning and the study of multi-agent systems are central research fields. This thesis focuses on a combination of these two areas: that is, a specialized kind of planning known as Multi-agent Epistemic Planning. This field of research is concentrated on all those scenarios where agents, reasoning in the space of knowledge/beliefs, try to find a plan to reach a desirable state from a starting one. This requires agents able to reason about her/his and others’ knowledge/beliefs and, therefore, capable of performing epistemic reasoning. Being aware of the information flows and the others’ states of mind is, in fact, a key aspect in several planning situations. That is why developing autonomous agents, that can reason considering the perspectives of their peers, is paramount to model a variety of real-world domains. The objective of our work is to formalize an environment where a complete characterization of the agents’ knowledge/beliefs interactions and updates are possible. In particular, we achieved such a goal by defining a new action-based language for Multi-agent Epistemic Planning and implementing epistemic planners based on it. These solvers, flexible enough to reason about various domains and different nuances of knowledge/belief update, can provide a solid base for further research on epistemic reasoning or real-base applications. This dissertation also proposes the design of a more general epistemic planning architecture. This architecture, following famous cognitive theories, tries to emulate some characteristics of the human decision-making process. In particular, we envisioned a system composed of several solving processes, each one with its own trade-off between efficiency and correctness, which are arbitrated by a meta-cognitive module.