The success of CMUnited-98 at RoboCup-98 was due to several technical innovations ranging from predictive locally optimal skills (PLOS) to strategic positioning using attraction and repulsion (SPAR). Building on the innovations of CMUnited-97, including flexible formation, a novel communication paradigm, and machine learning modules, CMUnited-98 successfully combines low-level individual and high-level strategic, collaborative reasoning in a single multi-agent architecture.
For a more thorough understanding of the implementation details involved, the reader is encouraged to study the algorithms described here in conjunction with the CMUnited-98 source code [11]. Other RoboCup researchers and multi-agent researchers in general should be able to benefit and build from the innovations represented therein.