DISCUSSION PROPOSAL: Technology Transfer Applying Mobile Robotics Research to Computer Games Brian Yamauchi Navy Center for Applied Research in Artificial Intelligence Naval Research Laboratory Washington, DC 20375-5337 yamauchi@aic.nrl.navy.mil Introduction Mobile robots and computer game AI systems are required to address many of the same problems. Both need the ability to: - Navigate in dynamic environments - React in real-time to unpredictable events - Plan in a spatial domain - Learn from experience - Cooperate with other robots, simulated agents, or human beings Recent advances in robotics research have led to the development of mobile robots that can perform increasingly complex tasks: - Exploration systems have been combined with robust navigation algorithms, enabling mobile robots to learn maps as they navigate through unknown territory. [7][9] - Behavior-based control systems have been combined with map-based and symbolic planning systems. These hybrid architectures allow robots to make plans based on a world model while also reacting quickly to unexpected events. [1] - A wide variety of algorithms have enabled robots to learn how to act in the world -- including neural networks, genetic algorithms, and reinforcement learning. [4][5] - Coordination strategies allow multiple robots to cooperate on complex tasks such as exploration and team sports. [3][8] Given the overlap between robotics and game AI tasks, how can the results from robotics research be applied to create better games? It may be useful to consider the ways in which game AI tasks are easier than robotics tasks -- and also the ways in which they are more difficult. For example, perception is much more difficult in the real world. In a game, it is trivial to give each agent complete knowledge of the simulated world. However, this approach is likely to lead to unrealistic behavior on the part of the game agents. If monsters can see through walls in an action game, or if the computer player has complete knowledge of a player's forces in a strategy game, then the player may feel that the computer is cheating, and have a less enjoyable experience. Limiting the knowledge available to each agent may make the game more believable. Such limitations require agents to deal with many of the problems faced by robots. If a predator's perception in an action game is limited to what it can see and hear, then it will have to actively search the environment for its prey. In this case, robot exploration and navigation techniques could prove useful. Likewise, if a computer adversary in a strategy game is limited in its knowledge of the player's forces, it may be able to apply some techniques that allow robots to deal with uncertain knowledge about the world. In the past, a factor that made game AI development more difficult was the limited computing power available on personal computers. In contrast, today's home PCs have power equivalent to that found on many state-of-the-art research robots. (Each of the Nomad 200 mobile robots used in our research at NRL uses a 133MHz Pentium chip as its primary processor.) The power now available for home PCs may make the application of advanced techniques from robotics more feasible, though speed and computational cost will continue to be important issues. In addition to applications to conventional action and strategy games, AI and robotics technologies also allow the development of new forms of entertainment. Examples include the work at the Oz Project [2] and Zoesis on interactive drama and at PF.Magic on virtual pets [6]. These examples combine AI and robotics techniques with insights from drama, film, and conventional animation in order to create engaging, believable characters. For future games, an important issue is how best to conduct the highly interdisciplinary work of developing these new forms of entertainment. Questions for Discussion - What sorts of robotics/AI techniques can be applied to control agents in action games -- behavior-based robotics, navigation, exploration, learning? - What sorts of robotics/AI techniques can be applied to adversary intelligence for strategy games -- planning, spatial reasoning, learning? - What other sorts of games can benefit from the application of robotics/AI techniques? - How can techniques developed for multirobot cooperation be applied to multiplayer games involving multiple human players and one or more AI players? - Can offline learning techniques (neural nets, genetic algorithms, Q-learning) be used to generate AI systems for controlling game agents? - What forms of perception (e.g. vision, hearing, smell) can be simulated effectively in order to create more realistic game agent behavior? - How much computing power will be allocated to game AI (as opposed to graphics, for example)? - What sort of robotics/AI techniques can respond in real-time using the available computing power? - How can robotics/AI techniques be used to create characters with that display distinct personalities and engage users on an emotional level? - What new forms of entertainment do advances in robotics and AI technology make possible? - What is the best way to develop these new forms of entertainment, when this development may require the combined efforts of AI and robotics researchers, game designers and programmers, and many different types of artists? References [1] Bonasso, Peter; Firby, James; Gat, Erann; Kortenkamp, David; Miller, David; Slack, Marc, "Experiences with an Architecture for Intelligent, Reactive Agents, Journal of Experimental and Theoretical Artificial Intelligence, January 1997. [2] Loyall, A. Brian; and Bates, Joseph, "Real-Time Control of Animated Broad Agents," Proceedings of the Fifteenth Annual Conference of the Cognitive Science Society, Boulder, CO, June 1993. [3] Noda, Itsuki; Suzuki, Shoji; Matsubara, Hitoshi; Asada, Minoru; and Kitano, Hiroaki, "RoboCup-97: The First Robot World Cup Soccer Games and Conferences," AI Magazine, Fall 1998, pp. 49-59. [4] Pomerleau, Dean, Neural Network Perception for Mobile Robot Guidance, Kluwer Academic Publishing, 1993. [5] Schultz, Alan; Grefenstette, John; and Adams, William, "Robo-Shepherd: Learning Complex Robotic Behaviors," Proceedings of the International Symposium on Robotics and Automation, May 1996, pp. 763-768. [6] Stern, Andrew; Frank, Adam; and Resner, Ben, "Virtual Petz: A Hybrid Approach to Creating Autonomous Lifelike Dogz and Catz," Proceedings of the Second International Conference on Autonomous Agents, Minneapolis, MN, May 1998, pp. 334-335. [7] Thrun, Sebastian; Fox, Dieter; and Burgard, Wolfram, "Probabilistic Mapping of an Environment by a Mobile Robot," Proceedings of the 1998 IEEE International Conference on Robotics and Automation, Leuven, Belgium, May 1998, pp. 1546-1551. [8] Yamauchi, Brian, "Frontier-Based Exploration Using Multiple Robots," Proceedings of the Second International Conference on Autonomous Agents, Minneapolis, MN, May 1998, pp. 47-53. [9] Yamauchi, Brian; Schultz, Alan; and Adams, William, "Mobile Robot Exploration and Map-Building with Continuous Localization," Proceedings of the 1998 IEEE International Conference on Robotics and Automation, Leuven, Belgium, May 1998, pp. 3715-3720.