petflyaiBack to the fly

SCIENCE & SOURCES

A body of knowledge.
One little life.

A fly is not just a shape that moves. It is a body, a nervous system, a history of encounters. We are bringing a growing body of open data and research into one simulated animal.

The aim is a connected loop: stimulus, sensation, neural activity, physical movement, and feedback. The prototype does not yet reproduce that full biological loop.

USED IN THE PROTOTYPE

The wiring is real.
The model is still a beginning.

Applied data · MaleCNS v1.0

From measured connections to a working circuit.

The current prototype uses a small, selected part of the male Drosophila central nervous system connectome. Original cell IDs, connection topology and synapse counts are retained; normalized weights carry activity through that circuit.

This is not a whole-brain simulation. Signal decay, thresholds, stimulus mapping and the interpretation of motor activity are authored approximations, not measured physiology. Food and water currently share a generic gustatory pathway.

Annotated cells
44
Weighted connections
99
  1. MeasuredSelected neural wiring

    MaleCNS cell identities and synapse counts.

  2. ApproximatedActivity propagation

    An engineered, positive-weight circuit model.

  3. InterpretedMovement signals

    Readouts influence authored body control.

MaleCNS is a collaboration of FlyEM at HHMI Janelia, the University of Cambridge, the MRC Laboratory of Molecular Biology, and Google Research. Data: CC BY 4.0. Our selected and normalized derivative is not endorsed by the source institutions.

Explore the datasetRelease historyInspect our circuit (JSON)

RESEARCH REFERENCES

An animal, studied
from many directions.

These sources guide anatomy, mechanics and the tests we want to reproduce. Citing a study does not mean its model or results are implemented in the prototype.

ANATOMY

The structures that make a fly.

An external anatomy atlas supplies regional references for eyes, mouthparts, wings, bristles and feet. Prepared microscope samples reveal form—not the color or material properties of a living fly.

Read the anatomy atlas

ARTICULATION

Joints, not convenient pivots.

CT-derived joint landmarks help ground how legs articulate. A measured specimen and fitted motion ranges are references, not universal joint limits or a complete muscle model.

Read the joint-axis study

BODY & FEEDBACK

A body the nervous system can act through.

NeuroMechFly and FlyGym inform our evaluation of articulated bodies, contact and sensory feedback. The current app uses custom reduced mechanics, not the full FlyGym runtime.

Explore FlyGym

NEURAL DYNAMICS

Wiring is a starting point, not a mind.

Shiu and colleagues' whole-brain model provides a reference for testing connectome-derived activity against selected experiments. Its demonstrated tasks do not establish complete fly physiology.

Read the neural-model study

BEHAVIOR

Grooming with a reason.

Seeds and colleagues studied how competing grooming actions are organized. We use this to ask what should trigger a sequence and what should stop it—not to treat an animation loop as biology.

Read the grooming study

INDIVIDUALITY

A tendency that belongs to this fly.

Buchanan and colleagues documented persistent turning preferences in repeated assays. This is a bounded reference for individual variation, distinct from thirst, learned responses or seeded randomness.

Read the study & explore its data

UNDER EVALUATION

A larger nervous system.
A question worth testing.

External demo · Not integrated

Xenova's browser-based neural simulation

The demo reports 166,700 MaleCNS neurons, locally simulated with custom WebGPU kernels. It is a promising reference for larger neural computation in the browser.

Its own disclosure matters: activity dynamics are simplified, and walking, turning and flight are crafted animations. That is not yet the closed sensory–neural–physical loop we are working toward.

Explore the demoRead the creator's post