Stone’s paper discusses the periodic team synchronization (PTS) domains as time-critical environments in which agents act autonomously with low communication, but in which they can periodically synchronize in a full-communication setting. Homogenous agents can flexibly switch roles within formations and agents can change formations dynamically, according to predefined triggers to be evaluated at run-time. This multi-agent system which involves several agents that collaborate towards the achievement of joint objective is viewed as a team of agents. Within this team member agent architecture and using locker-room agreement, we define out flexible teamwork structure that allows for task decomposition and dynamic role assignment in PTS domains.

This architecture is nice in that it deals with the issue of how to exchange messages between the different robots on a system domain with single-channel, low-bandwidth, unreliable communication. The communication between the players only has to communicate the type of formation which was already given to the players via the locker-room agreement. The messages themselves need some authentication measures because opposing players can replay messages to desynchronize the other team. The method to remedy this issue is by having the agents themselves independently keep track of the score and time. This timestamp is used to verify when the message came and to defeat any replay attacks.

This architecture also allows a flexible teamwork structure which improves over rigid structure by way of three characteristics: flexible positioning within roles, set-plays and changeable formations. In order to test the flexible teamwork structure, the team used a ball-dependent flexible position with set-plays against one using rigid positions and no set-plays. This is a nice approach to the dynamics of soccer because the agents can adapt their style of play depending on the opposing team and the current score. The approach of the article had a strong offense when losing and a strong defense if winning. This is a very smart approach to the game because of how quick a game can end with only one goal scored.

The team did very well during the competition with this strategy. The only team that they really lost to exploited vulnerability in the physics of the ball that was on that server. The issue was resolved after the competition but did not take away from the test of a flexible, dynamic system which allowed the robots to outscore their opponents 67-9. This approach demonstrated how one could manage numerous autonomous agents in an ever changing environment and successfully fulfill their goals at the same time.

Reference:

  1. Stone, Peter. “Task Decomposition, Dynamic Role Assignment, and Low-Bandwidth Communication for Real-Time Strategic Teamwork.” <http://robotics.usc.edu/~maja/teaching/cs584/papers/stone.pdf>