Execution of complex robotic tasks often requires efficient coordination of autonomic agents, which can be analyzed in a game-theoretic framework. Strategic, tactical and low-level control tasks have to be solved to efficiently coordinate mobile robots.
The main goal of this work is to develop such strategic methods that enable a set of agents to conquer and secure a certain area on a given map.
In my work, I examined the solution of efficient agent coordination for strategic and tactical level control tasks in the framework of reinforcement learning. The proposed solution extends reinforcement learning techniques from the set of single agent reinforcement learning methods, and a multi layer learning structure is defined which enables reinforcement learning in multiagent systems. The proposed solution should be capable of planning maneuvers for a team of agents to conquer or secure an area on a map.
The other focus of the thesis is the development of a framework which enables simulating and testing of multiagent coordination algorithms, and this framework is optionally capable of displaying the simulations in a physical environment.