CE33 - Interaction, Robotique – Intelligence artificielle

GRoups' Analysis for automated Cohesion Estimation – GRACE

GRACE:GRoups’ Analysis for automated Cohesion Estimation

Groups are a fascinating phenomenon. A 60 years long tradition on the study of groups in Social Science enabled to focus on higher level concepts called group emergent states. Cohesion is one of them. Despite cohesion being studied in several disciplines, a comprehensive multidisciplinary investigation is lacking. GRACE is a research project aimed at developing a computational model of cohesion integrating its task and social dimensions and also accounting for their relationship over time.

Mission and objectives

The project encompasses the following scientific, technological, and community-building objectives:<br /><br />Scientific objectives: to gain a deeper understanding of cohesion and, in particular, of the structural and temporal relationship between its major components, that is task and social components.<br /><br />Technological objectives: to investigate suitable technological solutions for collecting multimodal data from small groups. GRACE will exploit several sensing platform settings by looking at the recent improvements in wearable motion-capture technology and computer-vision based algorithms for multiparty detection and tracking; to develop software modules for automatically estimating cohesion and its components, accounting also for the temporal dimension.<br /><br />Community-building objectives: to improve scientific exchanges among researchers with the final aim to contribute to build an interdisciplinary scientific community working on emergent states and sharing the same research questions and methodological workflows.

First, GRACE performed a multidisciplinary critical review of the literature on cohesion to identify what is missing and the current open challenges. The outcomes of this first step allowed researchers to determine the theoretical framework and the evaluation tools their computational model have to ground on. This knowledge was then concretely applied to the design of an experimental scenario aimed at eliciting controlled changes of cohesion (increases/decreases). The chosen scenario was an escape game. A multimodal (motion capture, audio, video) data collection was performed on this scenario. Starting from this data and data publicly available, GRACE carried out several analysis aimed at identifying which multimodal features are the most relevant to cohesion and to set up a first baseline computational model.

Three remarkable results were obtained:
1) the GAME-ON dataset;
2) several scientific publications: 1 international journal paper, 2 international conference papers, and 1 national conference poster;
3) international collaborations to build an international research community working on similar/ complementary topics. This is witnessed by the scientific publications that include among the authors researchers from Germany and Italy, and by the organization of 3 workshops at international conferences in collaboration with researchers from Germany, Italy, the Netherlands, USA and Canada.

GRACE will produce a relevant scientific breakthrough by providing solid foundations from experimental evidence for developing models and algorithms for automated estimation of cohesion. The knowledge produced by GRACE will also be useful to identify novel requirements for a future generation of software applications able to provide feedback on group and team processes in several scenarios (e.g. meetings, surgery, education, sport training). Furthermore, GRACE will foster scientific exchanges among researchers from different European countries and working on different domains ranging from sociology to computer science. The obtained results may both open new and sustainable market opportunities and increase the competitiveness of companies in the area of Social Signal Processing in terms of novel products (e.g. software applications and social artificial agents) and services (e.g., providing technological support in highly cooperative scenarios).

1. Maman, L., Ceccaldi, E., Lehmann-Willenbrock, N., Likforman-Sulem, L., Chetouani, M., Volpe, G., and Varni. GAME-ON: A Multimodal Dataset for Cohesion and Group Analysis. IEEE Access, 8, 2020. ieeexplore.ieee.org/abstract/document/9127943

2. Ceccaldi, E., Lehmann-Willenbrock, N., Volta, E., Chetouani, M., Volpe, G., and Varni, G. How unitizing affects annotation of cohesion. In Proceedings of 2019 International Conference on Affective Computing and Intelligent Interaction (ACII). ieeexplore.ieee.org/abstract/document/8925527

3. F. Walocha, L. Maman, M. Chetouani, G.Varni. Modeling Dynamics of Task and Social Cohesion from the Group Perspective Using Nonverbal Motion Capture-based Features. Workshop IGTD, Proceedings of ACM ICMI 2020.

4. L. Maman. Multimodal Groups' Analysis for Automated Cohesion Estimation. Doctoral Consortium of ACM ICMI 2020, Proceedings of ACM ICMI 2020.

5. Maman, L. and Varni, G. GRACE: Un projet portant sur l’étude automatique de la cohésion dans les petits groupes d’humains. In Proceedings of Workshop sur les Affects, Compagnons Artificiels et Interactions (WACAI), 2020. hal.inria.fr/hal-02933474/document

GRACE is a fundamental research project contributing to axis 3 “Interaction, Robotique - Intelligence Artificielle” of challenge 7 “Société de l’information et de la communication”. The project aims at developing a computational model of cohesion among humans able to integrate the task and social dimensions of cohesion and also accounting for their relationship and their development over time. The model will be fed with multimodal nonverbal descriptors of cohesion computed at individual as well as group’s level. The impact of GRACE is expected in terms of i) requirements for a new generation of software applications capable of providing feedback on group processes (e.g. in meetings, surgery); ii) endowing artificial agents (e.g. virtual agents, robots) with skills to monitor and trigger cooperative behaviors both in everyday activities and in specialized tasks. This will open new market opportunities and increase competitiveness of companies in the area of social signal processing.

Project coordination

Giovanna VARNI (Laboratoire Traitement et Communication de l'Information)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.


LTCI Laboratoire Traitement et Communication de l'Information

Help of the ANR 219,564 euros
Beginning and duration of the scientific project: March 2019 - 36 Months

Useful links

Explorez notre base de projets financés



ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter