In: Soccer Robotics, pages 141-204. Volume 11 of Springer Tracts in Advanced Robotics --- Springer-Verlag, August 2004.
Abstract: The field of robot soccer provides numerous opportunities for the application of AI methods for game strategy development. As mentioned in the previous chapter, good strategies are needed to decide the roles and actions of team robots during the game. Chapter 4 has introduced a hybrid control architecture in which these strategies can be organized or integrated for proper management and control. In general, building a proper strategy is best guided by the intelligence aspects of search and evolution, knowledge representation and inference and learning and adaptation. In this chapter, these aspects of intelligence as needed by the DECIDE and ACT primitives and their importance are first discussed. The basics of some widely known soft-computing paradigms that make concrete (at least one of) these abstract aspects are then introduced. They include the formalisms of Petri nets, Q-learning, neural networks, evolutionary programming and fuzzy logic. Along which, the use of each paradigm for formulating strategies in robot soccer is motivated through simplified examples taken from previous FIRA Cup MiroSoT teams that demonstrate and emphasize its applicability in control, either at high-level (also called supervisory) or low-level. More speci.cally, for each paradigm, one or two examples are provided that address some key issues at specific hierarchical levels of the hybrid control architecture introduced in Chapter 4. By this, however, we do not imply that these paradigms cannot be applied at the other levels.
Back to the Petri Nets Bibliography