The field of cooperative multi-agent reinforcement learning (MARL) has seen widespread use in addressing complex coordination tasks. While value decomposition methods in MARL have been popular, they have limitations in solving tasks with non-monotonic returns, restricting their general application. Our work highlights the significance of joint intentions in cooperation, which can overcome non-monotonic problems and increase the interpretability of the learning process. To this end, we present a novel MARL method that leverages learnable joint intentions. Our method employs a hierarchical framework consisting of a joint intention policy and a behavior policy to formulate the optimal cooperative policy. The joint intentions are autonomously learned in a latent space through unsupervised learning and enable the method adaptable to different agent configurations. Our results demonstrate significant performance improvements in both the StarCraft micromanagement benchmark and challenging MAgent domains, showcasing the effectiveness of our method in learning meaningful joint intentions.

The responsibility framework was devised specifically for coordinating behavior in complex, unpredictable, and dynamic environments, such as industrial control. In distributed AI (DAI) systems, problem-solving agents cooperate to achieve the goals of the individuals and of the system as a whole. Each individual is capable of a range of identifiable problem-solving activities, has its own aims and objectives, and can communicate with others. Typically, agents within a given system have problem-solving expertise that is related but distinct and that has to be coordinated when solving problems. Such interactions are needed because of the dependencies between agents' actions and the necessity to meet global constraints and because often, no one individual has sufficient competence to solve the entire problem.

We describe a cognitive model of a collaborative agent that can serve as the basis for automated systems that must collaborate with other agents, including humans, to solve problems. This model builds on standard approaches to cognitive architecture and intelligent agency, as well as formal models of speech acts, joint intention, and intention recognition. The model is nonetheless intended for practical use in the development of collaborative systems.

My Ph.D. dissertation develops and implements a new model of multiagent coordination, called JOINT RESPONSIBILITY (Jennings 1992b), based on the notion of joint intentions. The responsibility framework was devised specifically for coordinating behavior in complex, unpredictable, and dynamic environments, such as industrial control. The need for such a principled model became apparent during the development and the application of a general-purpose cooperation framework (GRATE) to two real-world industrial applications.

My Ph.D. dissertation develops and implements a new model of multiagent coordination, called JOINT RESPONSIBILITY (Jennings 1992b), based on the notion of joint intentions. The responsibility framework was devised specifically for coordinating behavior in complex, unpredictable, and dynamic environments, such as industrial control. The need for such a principled model became apparent during the development and the application of a general-purpose cooperation framework (GRATE) to two real-world industrial applications.