SENSORIMOTOR PROCESSING BEYOND THE INDIVIDUAL: A VIRTUAL REALITY STUDY

In animal groups, sensorimotor algorithms for navigation must enable individual agents to coordinate and exchange information with one another. For example, in zebrafish, the trajectories of individuals at the front of the shoal ("leaders") may transmit relevant cues to other members ("followers"). However, the complexity of collective networks hinders our ability to establish causal information transfer. To experimentally control information flow between multiple individuals, we employed 3D immersive virtual reality (VR) for freely-swimming zebrafish. Each fish was hosted inside a VR apparatus and presented with virtual "avatars" of selected conspecifics. This setup allowed us to propagate information across experimentally defined social transmission chains. In parallel, we simulated transmission chains with artificial agents that behaved according to different types of sensorimotor controllers and decision-making algorithms. We then quantified the amount of information lost at each step of the chains using machine learning approaches. The comparison between fish and artificial agents allowed us to infer the computations that may underlie leader-follower behaviors and evaluate them in terms of information transfer between conspecifics. Overall, this study demonstrates a new experimental framework for investigating individual sensorimotor algorithms in terms of collective information processing, advancing our understanding of the computations that the brain must implement in order to generate this widespread biological function.