DESCRIPTION :
This postdoctoral research focuses on the modelling and simulation of dense human crowds, i.e. crowds in which social and physical interactions between individuals are mixed. The study of the behaviour of dense crowds is an important issue in terms of crowd management and accident prevention. Simulation of dense crowds is an essential part of these studies, but modelling the physical interactions between individuals still poses many unexplored challenges. The proposed postdoctoral program aims to make significant progress in this direction, and more specifically, the research on dense crowds will spawn new classes of Artificial Intelligence models.
To explore this area, the VirtUs team has a unique set of field data that captures the behaviour of dense crowds as observed at music festivals. The dataset includes videos of the crowd at festivals, a few manually checked tracked trajectories of individuals in the crowd, as well as automatically tracked trajectories of a large number of people in the image. The dataset also includes full-body motion capture for equipped participants in the crowd. As a continuation of recent work on modelling dense crowds, this postdoc aims to exploit the existing dataset for modeling and simulation purposes.
Objectives
The main goal of the postdoc is to establish new approaches to crowd simulation that are capable of learning new crowd dynamics based on their video observation, and to establish new crowd analysis methods based on simulation to improve the understanding of dense crowd behaviours.
The main activity of the postdoc will be
* To explore new crowd data representations for machine learning approaches to crowd simulation
* To explore new models for data-driven crowd simulation
* To explore new metrics for the evaluation of crowd simulation and comparison with real-world data.
The postdoc is open in terms of approaches and paradigms for crowd simulation, including trajectory prediction using generative models, or imitation of crowd behaviour using reinforcement learning approaches. Candidates may establish their own approach in discussion with the management team.
Outcomes
We will aim to publish the results of this research in the best conferences and journals in the field, including IEEE CVPR or ACM SIGGRAPH, ACM TOG.
The postdoc opens up the possibility of applying for permanent research positions in the VirtUs team at the Inria centre at the University of Rennes.
Bibliography
Chatagnon, T., Olivier, A. H., Hoyet, L., Pettré, J., & Pontonnier, C. (2025). Classification of first recovery steps after quiet standing following external perturbation from different directions. Journal of Biomechanics, 112639.
van Toll, W., Braga, C., Solenthaler, B., & Pettré, J. (2020, October). Extreme-density crowd simulation: combining agents with smoothed particle hydrodynamics. In Proceedings of the 13th ACM SIGGRAPH Conference on Motion, Interaction and Games (pp. 1-10).
Gomez-Nogales, G., Prieto-Martin, M., Romero, C., Comino-Trinidad, M., Ramon-Prieto, P., Olivier, A. H., ... & Casas, D. (2024). Resolving Collisions in Dense 3D Crowd Animations. ACM Transactions on Graphics, 43(5), 1-14.
He, F., Yue, J., Zhu, J., Seyfried, A., Casas, D., Pettré, J., & Wang, H. (2025). Learning Extremely High Density Crowds as Active Matters. arXiv preprint arXiv:2503.12168.
Sundararaman, R., De Almeida Braga, C., Marchand, E., & Pettre, J. (2021). Tracking pedestrian heads in dense crowd. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 3865-3875).
Dang, H. T., Gaudou, B., & Verstaevel, N. (2024). A literature review of dense crowd simulation. Simulation Modelling Practice and Theory, 102955.
Van Toll, W., & Pettré, J. (2021, May). Algorithms for microscopic crowd simulation: Advancements in the 2010s. In Computer Graphics Forum (Vol. 40, No. 2, pp. 731-754).
Principales activités
Main activities (5 maximum) :
* Propose solutions for automatic modelling of dense crowds based on data
* Coding / developping
* Analysis, evaluation of results
* Bibliographics studies
Code d'emploi : Data Scientist (h/f)
Domaine professionnel actuel : IT R&D Professionals
Temps partiel / Temps plein : Plein temps
Type de contrat : Contrat à durée indéterminée (CDI)
Compétences : Intelligence Artificielle, Vision par Ordinateur, C ++ (Langage de Programmation), Infographie, Machine Learning, Motion Capture, Reconnaissance de Formes, Apprentissage par Renforcement, Deep Learning, Modelsim, Anglais, Motivation Personnelle, KS0ENZ1L6D92JC45IEL0, Algorithmes, Biomécanique, Collisions, Elaboration des Prévisions, Conception et Design Graphique, Animation Graphique, Paradigmes, Recherche Post-Doctorale, Simulations, Compétences de Modélisation, Métrique, Imitation, Animation de Personnages, Prévention des Accidents
Courriel :
julien.pettre@inria.fr
Téléphone :
0299847100
Type d'annonceur : Employeur direct
