A detailed understanding of crowd dynamics is a societal necessity, both for safety reasons (evacuation, panic movements in large gatherings) and for the design of public spaces (train stations, shopping malls). For physicists, this raises the question of active matter in interaction, with the added complexity of the pedestrian's ability to anticipate the future movements of other pedestrians. The goal of this internship is to design and study a model of pedestrian motion which couples the microscopic/operational level of the motion with the necessity of optimization/anticipation required at a larger scale. This will involve relating the agent-based models to some equivalent kinetic and hydrodynamic models, and developing the corresponding mean-field game.