Analysing Soccer Games with Clustering and Conceptors

Abstract

We present a new approach for identifying situations and behaviours, which we call moves, from soccer games in the 2D simulation league. Being able to identify key situations and behaviours are useful capabilities for analysing soccer matches, anticipating opponent behaviours to aid selection of appropriate tactics, and also as a prerequisite for automatic learning of behaviours and policies. To support a wide set of strategies, our goal is to identify situations from data, in an unsupervised way without making use of pre-defined soccer specific concepts such as "pass" or "dribble". The recurrent neural networks we use in our approach act as a high-dimensional projection of the recent history of a situation on the field. Similar situations, i.e., with similar histories, are found by clustering of network states. The same networks are also used to learn so-called conceptors, that are lower-dimensional manifolds that describe trajectories through a high-dimensional state space that enable situation-specific predictions from the same neural network. With the proposed approach, we can segment games into sequences of situations that are learnt in an unsupervised way, and learn conceptors that are useful for the prediction of the near future of the respective situation.

Engineering Features to Improve Pass Prediction in Soccer Simulation 2D Games

“…Similarly, [3] has evaluated the kicking distribution and action trajectories of players from records of games to analyze teams' offensive strategy. The positional features of the players and the ball extracted from the records of the game have been a source of data for the behavioral analysis of players, such as the proposed methods in these papers [4,5,6] extracted 46 positional features of players and ball in the field for trajectory analysis, formation detection and assessment of games' state respectively, without any further investigation of the effect of other features groups. All these works had applied machine learning algorithms to the records of the game that were extracted when the game was finished; therefore the observation of agents and the process of decision making remains unclear.…”

Scientific and Technological Challenges in RoboCup

“…Gabel et al (140) evaluated the game quality across the past 20 years, observing that although the first decade showed amazing progress, the second did not. Michael et al (141) proposed a new approach for identifying situations and behaviors. Their goal was to identify situations from data in an unsupervised way without making use of predefined soccer-specific concepts, such as passing or dribbling.…”

RoboCupSimData: A RoboCup soccer research dataset

“…The provided data are useful for various different tasks including imitation learning (e.g., Ben Amor et al, 2013), learning or testing of self-localization (e.g., Olson, 2000), predictive modeling of behavior, transfer learning and reinforcement learning (e.g., Taylor and Stone, 2009), and representation learning for time series data (Michael et al, 2018). The next sections describe the environment, robots, and data in more detail.…”