A Structural Language for Multilevel Dynamics in the Design of Robot Soccer Systems

Johnson, Jeffrey and Rossi, Ruggero (2021). A Structural Language for Multilevel Dynamics in the Design of Robot Soccer Systems. In: Proceedings of International Conference on Artificial Life and Robotics, 25 pp. 270–276.

DOI: https://doi.org/10.5954/ICAROB.2020.PS-3

Abstract

Relational structure is ubiquitous in complex systems but very hard to implement in machines. Traditionally relational structures were hand-crafted using logic-based methods including various relational approaches to pattern recognition. Today the hope is that machines will find relational structures automatically by techniques of deep learning. Both approaches require new methods for representing relational structure for dynamic complex multilevel systems. We use the platform of robot soccer to investigate these ideas. This paper follows a previous paper which presented new dynamic structures for evolving tactics and strategies in team robotics. Here the notation is extended to include structures of structures of structures. For example a red defender robot r1 may closely mark a blue attacker robot b2 to create a structure < r1, b2; Rclosely_mark>. This may be part of another structure < < r1, b2; Rclosely_mark>, b3; Rdefenders_dilemma > as another robot b3 joins in to change the relational structure. This approach is illustrated by a RoboCup simulation game. Our next step is to build a competitive player to show that the ideas are operational and may give tactical and strategic advantages.