This project aims to develop a foundation model of primate behavior that captures the biomechanics and neural control of complex movements. Using multiview video recordings of macaques performing dexterous tasks, the project will build models that reconstruct detailed limb kinematics and infer underlying biomechanical states such as joint angles and muscle dynamics. By combining multiview pose estimation with biomechanical modeling and simulation, the project seeks to generate anatomically meaningful representations of movement that go beyond conventional keypoint tracking. These models will support the study of naturalistic motor behavior and provide a framework for linking detailed movement dynamics with neural activity.

The collaboration with the IBL Core will focus on developing robust multiview pose estimation pipelines, integrating biomechanical constraints into model training, and establishing evaluation benchmarks for behavioral foundation models. The project will also explore how different behavioral representations, such as kinematic trajectories or muscle-driven simulations, relate to neural activity. The resulting tools, models, and analysis pipelines will be released as open-source resources to support the broader neuroscience community and advance the study of complex motor behavior in primates.