Many of the most notable advances in AI from recent years come from systems trained to optimize behavior in a limited task domain, such as language or board games. Designing artificial systems that engage with complex environments and can optimize different task objectives depending on context remains a challenge. It has been suggested that animals' facility with such sensorimotor objectives relies on hierarchical control systems and that engineering systems with these principles in mind will accelerate research in AI and robotics. Hierarchical feedback loops exist at all levels of animal motor systems, from the body’s smart mechanical feedback (‘preflexes’), through spinal proprioceptive feedback, to feedback based on predictive models in the brain. However, this very complexity that gives natural systems their robustness makes them hard to understand and model. Modeling systems with feedback at multiple levels requires data at each level of description to constrain the model. Recently, such multi-scale data has become available for the motor system that generates visually controlled head movements of the fly, Drosophila melanogaster. We propose to use this data to generate a controller that can reproduce fly head movements via a physics-based model of the head motor system. This system presents an unprecedented opportunity to understand the neural implementation of a hierarchical control system at a circuit level.