Interpersonal coordination between the motion of two individuals performing a joint task has been extensively studied over the past few decades (Schmidt and Turvey, 1994; Richardson et al., 2007; Oullier et al., 2008; Schmidt and Richardson, 2008; Marsh et al., 2009; Varlet et al., 2011; Walton et al., 2015; Słowiński et al., 2016); a recent example being that of the mirror game, presented as paradigmatic case of study where human participants (HP) imitate each other's movements in a pair (Noy et al., 2011). In general, multiplayer scenarios have been investigated less than those involving only two participants, because of practical problems in running the experiments and the lack of models accounting for movement coordination in human groups, in contrast to the numerous studies dealing with animal groups (Couzin et al., 2005; Nagy et al., 2010, 2013; Zienkiewicz et al., 2015).

In this methodological paper we present a novel computer-based set-up that we named “Chronos” (a tool to study synCHRONizatiOn and coordination in human ensembleS), which allows participants to perform a joint task from a distance, both in dyads and in groups. The platform extends the mirror game to multiplayer scenarios, allowing each subject in the group to run a serious computer game where s/he can move the position of an object on her/his computer screen and see traces of the objects moved by the other players. The software makes it possible to show on each screen only the traces of a designated subset of the players in the group (as decided by an Administrator), so that different interaction patterns among its members can be implemented. To prevent any form of social interaction when in the same environment, participants are visually separated by barriers and hear white noise (through headphones connected to their computers) isolating them from any external sound. Alternatively, players can join the group remotely. Therefore, subjects have no information on the identity of those they are interacting with and receive no direct visual or behavioral cues from other members in the group.

Previous existing results on multiplayer human coordination include studies on rocking chairs (Frank and Richardson, 2010; Richardson et al., 2012; Alderisio et al., 2016b), group synchronization of arm movements and respiratory rhythms (Codrons et al., 2014), music (Glowinski et al., 2013; Badino et al., 2014; Volpe et al., 2016), and sport activities (Wing and Woodburn, 1995; Yokoyama and Yamamoto, 2011). However, in these papers the features and the level of coordination are not explicitly correlated to the way the players interact (i.e., the structure of their connections), as all the subjects involved share direct visual and auditory coupling with all the others, and no other patterns are considered. Moreover, inevitable social interaction affects the level of coordination in the group (Healey et al., 2005; Kauffeld and Meyers, 2009; Passos et al., 2011; D'Ausilio et al., 2012; Duarte et al., 2012, 2013; Glowinski et al., 2013; Cardillo et al., 2014). Indeed, body movements, friendship relationships, shared feelings, particular affinities, and levels of hierarchy have a significant impact on how each individual in the ensemble chooses her/his preferred partner(s) to interact the most with (Baumeister and Leary, 1995; Mäs et al., 2010; Stark et al., 2013). By making it possible to implement different visual pairings and minimize social interactions, our computer-based architecture allows instead to assess the impact on the group coordination level of solely varying the structure of interconnections among its members, a phenomenon that has been suggested to be crucial in determining the level of coordination arising in a human group (Passos et al., 2011; Duarte et al., 2012, 2013; Alderisio et al., 2016c).

In addition, Chronos allows the trace of some objects on the players' screen to be moved by virtual players (VP) driven by a computational cognitive architecture which is described in the rest of the paper. Simulated agents make it possible to further analyze the mechanisms underlying human coordination, explore features that are not easily accessible in ordinary human interactions, and point out interesting aspects of the task that are not immediately obvious from experimental observations (Di Paolo et al., 2008; Froese and Di Paolo, 2010). In so doing, we take inspiration from the human dynamic clamp (HDC) originally introduced in Dumas et al. (2014) as a paradigm to control the interactions between a human and a surrogate (virtual agent). By changing the mathematical description of the latter and appropriately tuning its parameters, it is possible to explore different behaviors and tasks, as well as test hypotheses and shed light on how humans interact. In our work we extend the HDC to a multiplayer scenario, thus allowing virtual agents to interact within a group of people.

To illustrate and validate the use of the set-up, we present preliminary experimental results obtained in a group of 5 individuals performing a joint oscillatory task. Specifically, we use the platform to vary the interaction patterns between players and observe the effects of such variations on the coordination level. Rather than being exhaustive, the experiments are reported here to illustrate the capabilities of the new methodology implemented via the platform. Therefore, we leave to future publications a more thorough experimental confirmation of the preliminary findings reported here for illustrative purposes. We also demonstrate that the platform allows the deployment of virtual computer players driven by feedback control algorithms. Specifically, we run preliminary experiments where a VP is enabled to interact with either one or a group of 4 human players, showing that its presence has an effect on their coordination level. Again, these experiments are not meant to be exhaustive but are only included to demonstrate the ability of the novel set-up to run trials involving multiple virtual players (a feature that has not been presented anywhere else in other methods in the literature on movement coordination in groups).

The new platform we present is available for download from https://dibernardogroup.github.io/Chronos to every interested reader.

The proposed computer-based platform is a hardware/software set-up consisting of input/output devices, a centralized unit (server and client-adiministrator) processing data, broadcasting movement information to the various client-players and implementing virtual agents, and a Wi-Fi apparatus connecting all the components together. The central server unit receives position data from the client-players and broadcasts to each position data from a subset of the others, according to the desired structure of interconnections being implemented. For example, in a ring network each client-player will only receive position data from two neighboring client-players. The movements of each human agent are detected by a low-cost position sensor, and individuals interact with each other through their own personal computer, on whose screens the central unit broadcasts the appropriate position trajectories according to the assigned topology (visual interaction patterns). The central unit is also responsible for data management and storage.

The proposed set-up is shown in details in Figure 1 for the case of N human participants and M virtual agents, and described below in all its components (for more information on how to use the set-up and for download of the software, see https://dibernardogroup.github.io/Chronos and Section 1 of Supplementary Material).

In this work we presented an ad hoc novel computer-based set-up for investigating human coordination, both in dyads and in groups, and showed preliminary results on coordination in human ensembles in order to validate its effectiveness. The proposed set-up allows to remove the effects of social interactions among the players and to implement different structures of interconnections. In addition, it allows to deploy virtual agents in the group, thus opening the possibility of further investigating the mechanisms that underly human group coordination through an extension of the human dynamic clamp to multiplayer scenarios (Dumas et al., 2014).

We envisage that the computer set-up presented in this paper can be used in Social Psychology to elucidate what the effects of social interactions are in dyadic or group movement coordination. Indeed, joint action tasks might first be performed while allowing participants to share direct visual and auditory coupling (participants directly look at each other instead of the screen of their personal computers, and do not wear headphones so that they know who they are interacting with), and then while removing them (or vice versa). Moreover, since some of the players can be replaced with one or more virtual agents, our computer technology can also be exploited for the development of artificial agents able to merge and interact within a group of humans (Boucenna et al., 2016; Iqbal et al., 2016), both for recreational (Alac et al., 2011) and rehabilitation purposes (Zhai et al., 2015; Bono et al., 2016; Słowiński et al., 2017).

In order to illustrate the features and capabilities of Chronos, we applied the platform to some representative scenarios. Specifically, we validated the use of a virtual player as designed in Alderisio et al. (2016a) in a dyadic interaction task. We found that the behavior exhibited in terms of the metrics used in Section 3.2 by each dyad was the same for both HP-HP and HP-VP interaction. This suggests that the human players involved in our experiments did not change the way of interacting with their partner according to the nature of the latter. In particular, we observed that if a human participant to whom a follower role was assigned in duo interaction ended up leading her/his human partner (in spite of the instruction given), s/he did so also when interacting with a virtual leader (Dyad 1). On the other hand, if a human leader was successfully leading her/his human partner, s/he did so also in the interaction with a virtual follower (Dyad 2). Despite being interesting, such preliminary results are specific for the trials we performed here to validate Chronos, hence they call for more experiments in order to be confirmed and extended.

Moreover, we illustrated the possibility of implementing different interaction patterns in larger ensembles. We observed that, only in the case of undirected topologies, coordination levels in a human ensemble are affected by the specific structure of interconnections among group members when any form of direct visual, auditory or social interaction is removed, a result found also in Alderisio et al. (2016c) yet in the presence of visual and social cues. This leads to open questions on what topology has to be implemented in order to enhance synchronization in the group, what the effects of removing some connections are, and whether the presence of social interaction further increases coordination.

Also, we validated the deployment of virtual agents in a group, and observed that they can decrease coordination levels when acting as leaders. Higher values of group and dyadic synchronization indices were instead observed either when no virtual player was interacting within the human ensemble, or when it was following the motion of all the subjects. These results only suggest that virtual players can be used to vary the level of coordination in a human group, although further work is required to better understand this effect and its implications.

Despite these results being promising, our experiments were presented in this methodological paper mainly to show the features of the computer-based set-up we propose. Rather than being exhaustive, they only illustrate the capabilities of Chronos and the analysis that can be carried out, hence we leave to future publications a more thorough experimental confirmation of the preliminary findings reported here for illustrative purposes.

Some further extensions to our work include the possibility of implementing time-varying topologies to study the effects of dynamically adding/removing connections among interacting participants (Cardillo et al., 2014), and enabling the administrator to provide the players with social cues in real time, based on the quality of their performance (i.e., as measured by the group synchronization index). In addition, it is possible to implement new mathematical models (Snapp-Childs et al., 2011; Zhai et al., 2016) for the VP to perform as joint improviser with other virtual or human agents. Finally, we are exploring the possibility of extending Chronos over the Internet, where it is also necessary to deal with network latency issues.

Conceived and designed the experiments: FA, GF, MdB. Performed the experiments: FA, ML, GF. Analyzed the data: FA, GF. Contributed analysis tools: FA, GF. Developed the software: ML. Wrote the paper: FA, GF, MdB.