"The German National Research Center for Information Technology in Berlin is working on a prototype to demonstrate its vision of a futuristic computer game ... These qualities will transform games of mindless violence into exercises on strategy." The Sunday Times Meanwhile, the project has expanded to a gobal scale, and involves contributers from all over the world. [ Please note: The project has been discontinued as of May 31, 2005 and is superseded by the projects of the ii Labs. There won't be further updates to these pages. ] The project's goal was to develop a generic architecture for autonomously operating agents, like computer-guided characters/mobiles/items, within a complex computer-game environment. These agents must be able to find the right actions to pursue their given goals and adapt their behavior to new environments or opponents. But not only the actions of the individual agent have to be intelligent, agents should also be able to communicate and perform coordinated group actions. For a domain like computer games, we have to take many features into account that conventional research rarely tackles: Real Time: There is only little time for reasoning, and approximate solutions need to be available soon. It should also be possible to slice the necessary computation time into small pieces such that higher-priority code does not have to wait until the full completion of our agent computations. Dynamics: Computer games provide a highly dynamic world. If the simulated world changes during the agent's computations, it should be possible to incorporate relevant changes on the fly instead of restarting the computations. Open World and Incomplete Knowledge: The agent does not know how many objects of which type are in the world, and has no complete information about their specific properties unless they can be sensed. The so-called "closed-world assumption" is thus not applicable, and we need to represent and handle incomplete information. Complex World: The world features symbolic as well as numerical properties, the duration of actions must be taken into account, actions can happen in parallel, actions may fail or have synergistic effects, and much more. Given these properties, the main research topics of the project are: Planning: How can we model and represent actions, states, goals, etc.? Current state: Mainly done, ongoing refinement Search: How can we efficiently search for a plan that satisfies the agent's goals? Current state: Mainly done, ongoing refinement Multi-Agent Systems: How can agents cooperate/coordinate/communicate? Current state: Partly done Learning: How can agents learn about their environment and develop new skills? Current state: Not even started An overview can be found in the [Preface] of the [Documentation] section.

For questions, comments or suggestions, please contact us.

Last update:
March 24, 2005 by Alexander Nareyek