Action Selection via Learning Behavior Patterns in Multi-Robot Systems
Can Erdogan and Manuela Veloso
The RoboCup Small Size League robot soccer competitions have successfully taken place for thirteen years with autonomous systems where a combination of centralized perception and control, and distributed actuation takes place. In a given game, teams of five robots move at high speeds in a limited space, actuating a golf ball, aiming to score goals. Although the teams perform in a compelling way in principle, running pre-planned strategies, adapting in real-time to the adversarial teams is still a big challenge. In this paper, we introduce a representation that models the spatial and temporal data of a multi-robot system as instances of geometrical trajectory curves. We then explain how to model the behavior of a multi-robot system by implementing a variant of agglomerative hierarchical clustering. Next, we provide an algorithm that classifies a behavior concurrently as it occurs, with respect to a given clustering. Subsequently, we define an algorithm that autonomously generates counter tactics. We evaluate our work on logs from real games and in simulation.