Fig 2 - uploaded by Oliver Michael Cliff
Content may be subject to copyright.
Interaction-base diagrams. Arrows represent highest information transfer between players. MATLAB copper colormap is used to indicate the strength of transfer, varying smoothly from black (weakest) to bright copper (strongest). Example interactions: four arrows in the right diagram from Marlik’s central mid-fielder, positioned in the centre circle, to all Gliders’ defenders indicate that the defenders respond mostly to the central mid-fielder’s motion. – Oxsy’s wing forwards mostly respond to Gliders’ side defenders, while Oxsy’s centre-forward does not mostly respond to Gliders’ centre-backs; – Oxsy’s side defenders mostly respond to Gliders’ wing forwards, while Oxsy’s centre-backs do not mostly respond to Gliders’ centre-forward; – Marlik’s forwards mostly respond to Gliders’ central mid-fielder; – Marlik’s defenders mostly respond to Gliders’ side-wingers. Now we turn our attention to the information-base interaction diagrams 2a and 2b: – Gliders’ defenders mostly transfer information to Oxsy’s wing forwards, but not to their centre-forward; – practically every Oxsy’s player transfers infromation to Gliders’ centre-forward; – Gliders’ defenders mostly transfer information to Marlik’s centre-forward, but not to their wing-forwards; – Gliders’ centre-forward is transferred information from many Marlik’s players, but not from their side defenders; – Gliders’s wing forwards are tightly coupled with Marlik’s side defenders. Even such a brief analysis helps to point out that in the contest with Oxsy, Gliders have a problem with their centre-backs not actively checking the opponent’s centre- forward, but a similar problem also exists in Oxsy’s own defense. Not surprisingly, most goals are scored in these games through the centre and not via the wing attacks and crosses. In addition, it appears that a lot of Gliders’ motion is tuned to opponents’ central mid-fielder which highlights a high degree of redundancy that may need to be exploited. In the games against Marlik it is evident that the opponents central mid- fielder plays a key role in both defense and attack, which again presents an opportunity to exploit such an overload. At the same time, it appears that a lot of interactions occur on the flanks of Marlik’s defense (defenders mark forwards who try to evade), while Marlik’s wing forwards are not marked by Gliders’s side defenders.
Source publication
We present several novel methods quantifying dynamic interactions in simulated football games. These interactions are captured in directed networks that represent significant coupled dynamics, detected information-theoretically. The model-free approach measures information dynamics of both pair-wise players’ interactions as well as local tactical c...
Context in source publication
Context 1
... all the transfers to team X are added up, and the transfers from team X are sub- tracted away. When each of the transfers is conditioned on some other contributor W (e.g., all the dynamics are computed in the context of the ball movement), the overall tactical relative responsiveness ∆ ( X , Y | W ) is also placed in this specific context, W . In principle, competitive situations result in quite vigorous dynamics within the involved lines and overall formations, and the team that manages to achieve a higher degree of tactical relative responsiveness does often perform better. While this is not a hard rule, we may correlate the scores of relative responsiveness (e.g., line-by-line) with the game scores, and identify the lines which impacted on the games more. Our tactical analysis also involves computation of the active information storage within the teams. We characterise team’s rigidity A X as the average of information storage values for all players of the team. We also determine the relative rigidity A ( X , Y ) = A − A for the teams (or their coupled lines). The hypothesis here is that The average information storage, or rigidity A X , is high whenever one can predict the motion of some players based on the movements of their other teammates. In these cases, the players are not as independent of each other as a truly complex or swarm behavior would warrant, making the tactics less versatile. Obviously, this may be counter-productive, since an opponent team can counteract by only partially ob- serving the ‘rigid’ team’s dynamics, and deducing the rest. Consequently, the relative rigidity A ( X , Y ) should be anti-correlated with the team X performance against team Y . To compute the measures described in previous sections, produce interaction diagrams and correlate tactical responsiveness with team performance, we carried out multiple iterative experiments matching Gliders2013 up against some well-known teams, such as Oxsy [18] and Marlik [19]. The correlation scores (Pearson product-moment correlation coefficients) reported below were tested for statistical significance, and corrected for multiple comparisons. Figure 1 presents the information-sink interaction diagram D ( Oxsy , Gliders ) and the information-base interaction diagram D ˇ ( Oxsy , Gliders ) , built over almost 500 hundred games between Oxsy and Gliders. Analogously, Fig. 2 shows the information-sink interaction diagram D ˆ ( Marlik , Gliders ) and the information-base interaction diagram D ˇ ( Marlik , Gliders ) , built over nearly 450 hundred games between Marlik and Gliders. Several interesting observations can be made. In general, the diagrams are highly symmetric with respect to left and right wings. The diagrams represent interactions av- eraged over many games, and so the symmetry demonstrates that the employed methods are robust to noise present in individual games. Also, the information-sink diagrams do differ from information-base diagrams, as expected. We begin a more detailed analysis with the information-sink interaction diagrams 1a and ...
Similar publications
An important property of a telephony system is the call control model on which it is based. It is noted that many call control models in the past, especially those in PSTN/ISDN networks follow centralized model. For such a model, typical is significant coupling of modules belonging to different services with the basic call control module which is a...
The measured complete fusion (capture) excitation function is presented for the 28Si + 208Pb reaction at deep sub-barrier energies. This excitation function is compared with the one predicted with the quantum diffusion approach.
Aero-engine spline couplings are widely used in the aero-industry to transfer torque and power. They are often subjected to fretting wear damage which results in reduced equipment life and early replacement. This work focuses on experimental investigation to be used as a basis for predicting fretting wear in spline couplings working in a misaligned...
The quantum state transfer between two membranes in coupled cavities is studied when the system is surrounded by non-Markovian environments. An analytical approach for describing non-Markovian memory effects that impact on the state transfer between distant membranes is presented. We show that quantum state transfer can be implemented with high eff...
Citations
... Nonetheless, group systems, including team sports, are characterized by more than agents' position, often offering rich dynamic interactions able to shape the outcome (e.g., the end result of the match). In team sports, interactions between players of the same team are fostered with the intent to undermine the objectives of the opposing team, while fulfilling their own objectives (Cliff et al., 2013). This leads to complex spatio-temporal interactions, which may seem intractable to understand and quantify. ...
... Nonetheless, group systems, including team sports, are characterized by more than agents' position, often offering rich dynamic interactions able to shape the outcome (e.g., the end result of the match). In team sports, interactions between players of the same team are fostered with the intent to undermine the objectives of the opposing team, while fulfilling their own objectives (Cliff et al., 2013). This leads to complex spatio-temporal interactions, which may seem intractable to understand and quantify. ...
... In general, our results may have a broad appeal, emphasizing a continuing importance of the divide-and-conquer approaches to complex problems (Glasmachers, 2017). The structurepreserving imitation learning augmented by delayed rewards is likely to find applications in multi-agent cooperation and collective behavior (Prokopenko and Wang, 2004;Xu et al., 2013;Bai et al., 2015;Hamann et al., 2016;Cliff et al., 2017), modular robotics (Prokopenko et al., 2006;Martius et al., 2007;Der and Martius, 2012), and distributed networks and dynamical systems in general (Mortveit and Reidys, 2007;Cliff et al., 2013Cliff et al., , 2016Hefny et al., 2015). ...
We describe and evaluate a neural network-based architecture aimed to imitate and improve the performance of a fully autonomous soccer team in RoboCup Soccer 2D Simulation environment. The approach utilizes deep Q-network architecture for action determination and a deep neural network for parameter learning. The proposed solution is shown to be feasible for replacing a selected behavioral module in a well-established RoboCup base team, Gliders2d, in which behavioral modules have been evolved with human experts in the loop. Furthermore, we introduce an additional performance-correlated signal (a delayed reward signal), enabling a search for local maxima during a training phase. The extension is compared against a known benchmark. Finally, we investigate the extent to which preserving the structure of expert-designed behaviors affects the performance of a neural network-based solution.
... The methods of big data analytics could be exploited to analyze the data generated in football matches, the social network analysis is considered a very important one, and relevant studies are well documented [4][5][6][7][8][9][11][12][13] . However, there are several aspects that need to be improved, such as limited performance indicators considered, simplistic statistical methods used, and overly rigorous modeling. ...
... From a network perspective, football match presents a competitive relationship among dynamic network topology nodes. Cliff et al. [11] proposed a method of constructing interactive networks, which could reveal the significant coupling dynamics generated by team games or other activities with concurrent cooperation and competition characteristics. Using passing data, Peña et al. [22] built a weighted and directional network for each team, from which they could determine the game pattern, and to determine the hot spots in the match and detect potential weaknesses as well. ...
The current study constructed social network using player positions and passing process based on available literature. Multiple indicators were used to measure the importance of positions comprehensively. The results showed that in the football passing process, the attacking midfielder was the most important position, followed by the central defending midfielder. Based on two-sample difference tests, the results showed that the winning teams usually had better performance on positions of forward on left, central forward, defender on left and defender on right. To analyze the effects of playing positions on the whole network and test the sensitivity of passing networks, we deleted n (n = 1, 2, 3…, 10) positions of a team, and then tested the efficiency of the networks based on positions left.
... In general, the objective function and, thus the dynamic local constraints, may be chosen to represent different aspects of the problem structure: maximisation of spatiotemporal coordination [31,32], maximisation of information flows [47,48], maximisation of thermodynamic efficiency which contrasts the gain in the uncertainty reduction with the required thermodynamic work [49], etc. ...
Fractals2019 started as a new experimental entry in the RoboCup Soccer 2D Simulation League, based on Gliders2d code base, and advanced to a team winning RoboCup-2019 championship. Our approach is centred on combinatorial optimisation methods, within the framework of Guided Self-Organisation (GSO), with the search guided by local constraints. We present examples of several tactical tasks based on the fully released Gliders2d code (version v2), including the search for an optimal assignment of heterogeneous player types, as well as blocking behaviours, offside trap, and attacking formations. We propose a new method, Dynamic Constraint Annealing, for solving dynamic constraint satisfaction problems, and apply it to optimise thermodynamic potential of collective behaviours, under dynamically induced constraints.
... The latter comprises a set of evolutionary computation techniques that incorporate human innovation [21,22]. This fusion allowed us to optimise several components, including an action-dependent evaluation function proposed in Gliders2012 [23], a particle-swarm based self-localisation method and tactical interaction networks introduced in Gliders2013 [24,25,26,27,28], a new communication scheme and dynamic tactics with Voronoi diagrams utilised by Gliders2014 [29], bio-inspired flocking behaviour incorporated within Gliders2015 [30], and opponent modelling diversified in Gliders2016 [16]. The overall framework achieved a high level of tactical proficiency ensuring players' mobility and the overall control over the soccer field. ...
We describe Gliders2d, a base code release for Gliders, a soccer simulation team which won the RoboCup Soccer 2D Simulation League in 2016. We trace six evolutionary steps, each of which is encapsulated in a sequential change of the released code, from v1.1 to v1.6, starting from agent2d-3.1.1 (set as the baseline v1.0). These changes improve performance by adjusting the agents' stamina management, their pressing behaviour and the action-selection mechanism, as well as their positional choice in both attack and defense, and enabling riskier passes. The resultant behaviour, which is sufficiently generic to be applicable to physical robot teams, increases the players' mobility and achieves a better control of the field. The last presented version, Gliders2d-v1.6, approaches the strength of Gliders2013, and outperforms agent2d-3.1.1 by four goals per game on average. The sequential improvements demonstrate how the methodology of human-based evolutionary computation can markedly boost the overall performance with even a small number of controlled steps.
... Tracing such advances is especially important because different champion teams usually employ different approaches, often achieving a high degree of specialisation in a sub-field of AI, for example, automated hierarchical planning developed by WrightEagle [23,24,26,21,28], opponent modelling studied by HELIOS [27], and human-based evolutionary computation adopted by Gliders [11,12]. Many more research areas are likely to contribute towards improving the League, and several general research directions are recognised as particularly promising: nature-inspired collective intelligence [29,30,31], embodied intelligence [32,33,34,35], information theory of distributed cognitive systems [36,37,38,39,40,41], guided self-organisation [42,43,44], and deep learning [45,46,47]. ...
We summarise the results of RoboCup 2D Soccer Simulation League in 2016 (Leipzig), including the main competition and the evaluation round. The evaluation round held in Leipzig confirmed the strength of RoboCup-2015 champion (WrightEagle, i.e. WE2015) in the League, with only eventual finalists of 2016 competition capable of defeating WE2015. An extended, post-Leipzig, round-robin tournament which included the top 8 teams of 2016, as well as WE2015, with over 1000 games played for each pair, placed WE2015 third behind the champion team (Gliders2016) and the runner-up (HELIOS2016). This establishes WE2015 as a stable benchmark for the 2D Simulation League. We then contrast two ranking methods and suggest two options for future evaluation challenges. The first one, "The Champions Simulation League", is proposed to include 6 previous champions, directly competing against each other in a round-robin tournament, with the view to systematically trace the advancements in the League. The second proposal, "The Global Challenge", is aimed to increase the realism of the environmental conditions during the simulated games, by simulating specific features of different participating countries.
... planar) coordinates and their changes. The problem is difficult as some of the dependencies between agents are not discernible simply by correlating their corresponding locations over time — one needs to take into account the possibly directed nature of such correlations, where dynamics of one of the agents affects the positioning of another [11]. Quantitative analysis, in particular using complex systems theory, is increasingly being used in team sports to better understand and evaluate perfor- mance [1, 69] and identify networks between players. ...
... The contributions of this paper are three-fold. Firstly, we further validate our information-theoretic analysis of the RoboCup-2012 2D Simulation [11] with most recent data-sets from RoboCup-2014 and new relative measures. Secondly, we extend the analysis towards datasets that reveal asymmetric interaction networks ; which, for the first time, discern and quantify a number of asymmetries in the tactical schemes used by the teams. ...
... Table 1summarises different relative measures, specified for different tactical roles in a typical football forma-tion. We would like to point out that we introduce here new relative measures, expanding on the ones analysed in [11]. Specifically, the previous study [11] compared attacking vs defending lines, that is, analysed T Ya→X d − T X d →Ya , while in this work we compare attacking vs attacking lines on the one hand, ...
We develop and apply several novel methods quantifying dynamic multi-agent team interactions. These interactions are detected information-theoretically and captured in two ways: via (i) directed networks (interaction diagrams) representing significant coupled dynamics between pairs of agents, and (ii) state-space plots (coherence diagrams) showing coherent structures in Shannon information dynamics. This model-free analysis relates, on the one hand, the information transfer to responsiveness of the agents and the team, and, on the other hand, the information storage within the team to the team's rigidity and lack of tactical flexibility. The resultant interaction and coherence diagrams reveal implicit interactions, across teams, that may be spatially long-range. The analysis was verified with a statistically significant number of experiments (using simulated football games, produced during RoboCup 2D Simulation League matches), identifying the zones of the most intense competition, the extent and types of interactions, and the correlation between the strength of specific interactions and the results of the matches.
... Starting from 2013, Gliders utilised information dynamics [42,43,44,45,46,47] for tactical analysis and opponent modelling. This analysis involves computation of information transfer and storage, relating the information transfer to responsiveness of the players, and the information storage within a team to the team's rigidity and lack of tactical richness. ...
... We thank David Budden for developing a new self-localisation method introduced in Gliders2013 [49,34] as well as contributing to the analysis of competition formats [4], and Oliver Cliff for developing a new communication scheme adopted by Gliders from 2014 [35]. The overall effort has also benefited from the study quantifying tactical interaction networks, carried out in collaboration with Oliver Cliff, Joseph T. Lizier, X. Rosalind Wang and Oliver Obst [46]. We are thankful to Ivan Duong, Edward Moore and Jason Held for their contribution to Gliders2012 [33]. ...
We review disruptive innovations introduced in the RoboCup 2D Soccer Simulation League over the twenty years since its inception, and trace the progress of our champion team (Gliders). We conjecture that the League has been developing as an ecosystem shaped by diverse approaches taken by participating teams, increasing in its overall complexity. A common feature is that different champion teams succeeded in finding a way to decompose the enormous search-space of possible single- and multi-agent behaviours, by automating the exploration of the problem space with various techniques which accelerated the software development efforts. These methods included interactive debugging, machine learning, automated planning, and opponent modelling. The winning approach developed by Gliders is centred on human-based evolutionary computation which optimised several components such as an action-dependent evaluation function, dynamic tactics with Voronoi diagrams, information dynamics, and bio-inspired collective behaviour.
... Finally, we show that the log-likelihood and log-likelihood ratio can be expressed in terms of collective transfer entropy (Lizier et al., 2010;Vicente et al., 2011). This result places our work in the context of effective network analysis (Sporns et al., 2004;Park and Friston, 2013) based on information transfer (Honey et al., 2007;Lizier et al., 2011;Cliff et al., 2013Cliff et al., , 2016) and relates to the information processing intrinsic to distributed computation (Lizier et al., 2008). ...
... The results presented here provoke new insights into the concepts of structure learning, non-linear time series analysis, and effective network analysis (Sporns et al., 2004;Park and Friston, 2013) based on information transfer (Honey et al., 2007;Lizier et al., 2011;Cliff et al., 2013Cliff et al., , 2016. The information-theoretic interpretation of the log-likelihood has interesting consequences in the context of information dynamics and information thermodynamics of non-linear dynamical networks. ...
The behaviour of many real-world phenomena can be modelled by nonlinear dynamical systems whereby a latent system state is observed through a filter. We are interested in interacting subsystems of this form, which we model as a synchronous graph dynamical system. Specifically, we study the structure learning problem for dynamical systems coupled via a directed acyclic graph. Unlike established structure learning procedures that find locally maximum posterior probabilities of a network structure containing latent variables, our work exploits the properties of certain dynamical systems to compute globally optimal approximations of these distributions. We arrive at this result by the use of time delay embedding theorems. Taking an information-theoretic perspective, we show that the log-likelihood has an intuitive interpretation in terms of information transfer.
