First Full‑Brain Simulation of a Fruit Fly Brings Brain‑Upload Closer to Reality

What Happened

In March 2026 Eon Systems demonstrated a complete digital reconstruction of a fruit‑fly brain containing ~125 000 neurons and ~50 million synapses, driving a physically simulated body to walk, turn and take off.

Key Technical Steps

Acquisition of the full fruit‑fly connectome (every neuron and synapse).

1:1 recreation of the network in software (125 k neurons, 50 M synapses).

Integration of the digital brain with a physics‑based body in a closed‑loop sensor‑neural‑actuator cycle.

FlyWire Connectome

The connectome was produced over ten years (FlyWire project). Researchers sliced a fruit‑fly brain into 7 000 ultra‑thin sections, imaged each with electron microscopy, and used AI‑assisted annotation to trace neuron morphology and synaptic connections. The resulting map was published in Nature (Oct 2024) with Philip Shiu as a lead author. In that paper the computational model achieved 95 % accuracy in predicting fruit‑fly locomotion.

Simulation Stack and Closed‑Loop Operation

The demonstration combined Shiu’s whole‑brain model, the NeuroMechFly v2 simulation framework, and the MuJoCo physics engine. The loop operated as:

sensory input → full‑connectome signal propagation → motor command output → physical body execution

Observed Autonomous Behaviors

Gait initiation : stable walking achieved within ~80 ms from rest.

Straight‑line walking : three‑leg coordination at ~3 cm/s.

Turning : asymmetric stride modulation producing ~10 rad/s curved turns.

Take‑off : controlled flight at ~20 cm/s.

These behaviors outperformed control networks built from random graph structures, indicating that the biological wiring encodes efficient motor control.

Comparison with Earlier Whole‑Brain Simulations

OpenWorm (2014) : simulated 302‑neuron C. elegans, no vision, limited behavior.

DeepMind / Janelia virtual fruit fly (2024) : used reinforcement learning to train a neural network that can fly; the “brain” consisted of learned weights, not a measured connectome.

Eon Systems (2026) : employed a measured 125 k‑neuron connectome, driving a physics‑based body in a closed loop—the first full chain from biological connectome to autonomous behavior.

Model Simplifications and Controversy

The neuron model uses a leaky integrate‑and‑fire (LIF) scheme, assuming uniform excitatory dynamics and ignoring morphological differences, glial cells, and chemical modulation. Critics argue that a static connectome alone cannot capture full brain function. Supporters point to the 95 % motion‑prediction accuracy as evidence that connectivity topology carries substantial functional information.

Implications

Neuro‑disease research : a digital brain can be perturbed to model Parkinson’s, Alzheimer’s, etc., enabling in‑silico drug testing.

Brain‑inspired AI : the efficiency of the biological wiring may inform next‑generation AI architectures.

Neuroscience tool : arbitrary neurons can be manipulated in the simulation, allowing experiments not feasible in vivo.

Illustrations