Predicting the Movements of Robot Teams Using Generative Models
When a robot plans its actions within an environment containing multiple robots, it is often necessary to take into account the actions and movements of the other robots to either avoid, counter, or cooperate with them, depending on the scenario. Our predictive system is based on the biologically-inspired, simulationtheoretic approach that uses internal generative models in single-robot applications. Here, we move beyond the single-robot case to illustrate how these generative models can predict the movements of the opponent’s robots, when applied to an adversarial scenario involving two robot teams. The system is able to recognise whether the robots are attacking or defending, and the formation they are moving in. It can then predict their future movements based on the recognised model. The results confirm that the speed of recognition and the accuracy of prediction depend on how well the models match the robots’ observed behaviour.
KeywordsMultiagent System Internal Model Inverse Model World State Predictive System
Unable to display preview. Download preview PDF.
- Beetz, M., Kirchlechner, B.: Computerized real-time analysis of football games. IEEE pervasive computing 4(3) (2005)Google Scholar
- Darken, R., Mcdowell, P., Johnson, E.: The Delta3D open source game engine. IEEE computer graphics and applications 25(3) (2005)Google Scholar
- Demiris, Y., Khadhouri, B.: Hierarchical attentive multiple models for execution and recognition of actions. Robotics and autonomous systems 54(5) (2006)Google Scholar
- Devaney, M., Ram, A.: Needles in a haystack: Plan recognition in large spatial domains involving multiple agents. In: National Conference on Artificial Intelligence (1998)Google Scholar
- Kitano, H., Asada, M., Kuniyoshi, Y., Noda, I., Osawai, E., Matsubara, H.: Robocup: A challenge problem for AI and robotics. In: Kitano, H. (ed.) RoboCup 1997. LNCS, vol. 1395, pp. 1–19. Springer, Heidelberg (1998)Google Scholar
- Takács, B., Butler, S., Demiris, Y.: Multi-agent behaviour segmentation via spectral clustering. In: Proceedings of the AAAI-2007, PAIR Workshop, pp. 74–81. AAAI Press, Menlo Park (2007)Google Scholar