A general framework for anticipatory agents based on the concept of anticipatory systems has been developed. The general idea is that each agent should have an internal model of its environment (including itself) and by running this model faster than real time make predictions of future states. These predictions are then used to adapt the agent's behavior. As a specialization of this idea, I have developed linearly anticipatory agent architecture. Some promising results on both competitive and collaborative behavior in multi-agent systems using this architecture have been presented.
Contact Paul Davidsson
Back to Projects