A Complete Fruit Fly Brain Simulation Now Controls a Virtual Body

Eon Systems has achieved a significant milestone in computational neuroscience by demonstrating a complete, biologically-accurate fruit fly brain simulation that controls a physics-based virtual body in a closed-loop system. The model is based on the FlyWire connectome—a detailed map of neural wiring—and contains over 125,000 neurons and 50 million synaptic connections. Led by senior scientist Philip Shiu, the team used machine learning to predict neurotransmitter identities, creating a functional digital twin of the Drosophila melanogaster brain. This work moves beyond static anatomical maps into the realm of interactive digital organisms.

Previously, the 2024 brain model was a brain without a body, capable of simulating activity but not interacting with an environment. The new demonstration bridges this gap by connecting the neural model to a simulated fly body using the MuJoCo physics engine, commonly used in robotics. The virtual fly now exhibits complex, emergent behaviors like walking, grooming, and feeding. Critically, these actions are not pre-programmed animations but arise from the brain model's own neural circuits processing simulated sensory input, such as the presence of virtual sugar, and generating appropriate motor outputs.

This integration represents a unique synthesis of two previously separate research tracks: detailed connectome mapping and realistic embodied simulation. It provides a powerful new platform for testing hypotheses about how specific neural circuits generate behavior, offering insights that could eventually inform the development of more efficient, brain-inspired AI architectures and advance our understanding of biological intelligence.