Research

Founding vision and positioning

The first living connectome applied to a synthetic/robotic body.

We keep the ambition of that idea and translate it into a defined scientific aim: to investigate the functional embodiment of a model derived from the MaleCNS in a physical robot, with closed-loop perception and action and reproducible results.

“Living connectome” is a metaphor for a running model interacting with its environment. The source material is a reconstructed anatomical map. There is no transfer of living tissue here, no brain transplant, no demonstration of consciousness and no continuity of a fly's identity.

Initial public positioning

openflymind is an initiative that aims for a pioneering demonstration of MaleCNS robotic embodiment, documenting the experimental path along the way.

Related work

• Connectome/GoPiGo[R7]runs a C. elegans model on a physical robot. It is a documented precedent of a connectome-based model connected to hardware.
• FLYNN[R8]uses the topology of the female Drosophila FAFB v783 connectome for simulated robot navigation in MuJoCo. It is not a demonstration on physical hardware and it does not use the MaleCNS.

The broad claim of being the first connectome of any living being applied to any robot is not published as fact. This initial research also does not establish priority within a narrower scope. Should a future review support a claim of first demonstration, the wording will state exactly which dataset, version, coverage, physical modality and experimental conditions are involved, and it will be published only together with the corresponding evidence.

Vocabulary

These are the terms we use throughout the site. Each has a fixed meaning in the project.

We avoid any wording that suggests living tissue, transferred consciousness, an implanted brain, human intelligence or a complete reproduction of the animal — the project does none of that.

Scientific basis

We use the FlyEM Male CNS Connectome, presented by HHMI Janelia and collaborators [R1]. It covers the central brain, the optic lobes and the ventral nerve cord of an adult male Drosophila melanogaster. The openflymind team builds on this resource and takes no credit for the original extraction.

These numbers describe the published resource, not what openflymind has imported or run.

The dataset is credited to FlyEM/HHMI Janelia, the University of Cambridge, the MRC Laboratory of Molecular Biology and Google Research [R1]. Use of public data does not imply institutional collaboration with openflymind.

Synaptic contacts, pre- and postsynaptic sites, neuron pairs and aggregated edges are not equivalent units. The 125 million figure stays attributed to the institutional announcement and is not converted into matrix dimensions or edge counts. Counts for each import will be computed with recorded filters and version.

What “translating into a matrix” means

In plain language, the proposal follows seven steps. None of them is complete; the list describes the method, not the current state.

1. Select and record the data. Identify the neurons, connections and annotations needed, with version, filters, provenance and checksum.
2. Build a directed graph. Nodes represent neurons or explicitly grouped units; edges represent connections present in the data.
3. Represent connectivity as sparse matrices. Keep traceability between computational indices and biological identifiers.
4. Define the neural dynamics. Choose update equations, scales, parameters and hypotheses. An anatomical matrix alone does not determine the full dynamics of a nervous system.
5. Connect sensors. Turn camera signals and, later, touch and hearing sensors into inputs compatible with the modelled populations.
6. Connect actuators. Map model outputs to movement controllers, with physical limits and a safe stop.
7. Close the loop and measure. The robot acts in the environment; new stimuli return to the sensors; experiments assess the behaviour.

Explanatory mathematical scheme

The scheme explains the idea; it is not an algorithm that has already been chosen.

A[i,j] = number of observed contacts from j to i
W      = documented transformation of A and of the model's hypotheses
x(t+1) = F(W x(t), B u(t), parameters)
a(t)   = G(C x(t))

Here u stands for sensory inputs, x for the model states and a for motor commands. The auxiliary functions and matrices depend on the chosen protocol.

We will record which weights are derived from anatomy, which are normalised and which, if any, are learned. Physiological strength and excitatory or inhibitory sign are not inferred automatically from synapse counts or neurotransmitter labels. Every simplification is treated as a traceable hypothesis.

Four layers that are never conflated

The project may start with selected circuits and widen its coverage over time. That is why we always distinguish four layers, each with its real state:

Questions and tests

This is the research agenda. The questions are open and there are no results for any of them yet.

1. How can features of biological connectivity be preserved when translating it into an executable model?
2. Which mapping lets artificial sensors feed circuits derived from this connectome?
3. How do the choice of body, sensors and dynamics change the observed behaviour?
4. What does biological topology contribute when compared with adequate controls?
5. Does the system keep stable responses under controlled changes to the environment or reduced sensory input?

Controls

Every experiment allows comparison with a conventional controller and, where relevant, with control graphs of comparable size and connectivity. Without those controls, an observed behaviour says nothing about the contribution of biological topology.

Experiment record

Each experiment published in the journal can record:

• task
• condition
• repetitions
• seed
• versions
• latency
• trajectory
• collisions
• interventions

No metric is imposed before it has been measured.

This page shows no results because there are none yet. When they exist, they will be published in the journal with method, conditions and limitations.

Milestones

The milestones below come from the same record that feeds the journal. Their states are real: a milestone changes state only when something has been recorded. We do not use completion percentages.

1. 01

   Foundations and scope

   In progress

   Selection of MaleCNS, formulation of the question and definition of evaluation criteria.

   Dataset chosen and question formulated; evaluation criteria being drafted. Decisions count as documented only once recorded in the journal.

2. 02

   Import and graph

   Planned

   Reproducible import artefacts (version, filters, provenance, checksum) and a sparse connectivity matrix.

3. 03

   Model dynamics

   Planned

   Implementation of the update equations with identified, traceable hypotheses.

4. 04

   Vision and movement in simulation

   Planned

   Experimental protocol and initial tests with simulated visual inputs and motor outputs.

5. 05

   Robotic body

   Planned

   Assembly of the physical platform and electronic integration of sensors, processing and actuators.

6. 06

   Physical closed loop

   Planned

   Experiments with real sensors and actuators, compared against appropriate controls.

7. 07

   Sensory extensions

   Planned

   Touch and hearing, subject to feasibility demonstrated in earlier stages.

   Exploratory.

8. 08

   Reproduction and communication

   Planned

   Code, methods and evidence made available for independent reproduction.

Follow the milestones in the journal

References

Scientific and data sources consulted for this page. Team and implementation references live on their own pages.

1. [R1]Janelia — Male CNS Connectome. HHMI Janelia Research Campus, FlyEM Project Team — Chosen resource: anatomical coverage, credits and licence.accessed on September 21, 2026
2. [R2]MaleCNS — male-cns.janelia.org — Versions, tools and release timeline.accessed on September 21, 2026
3. [R3]MaleCNS — Download — Programmatic access and data tables. The documentation consulted uses male-cns:v1.0.male-cns:v1.0 · accessed on September 21, 2026
4. [R4]Sexual dimorphism in the complete connectome of the Drosophila male central nervous system. Berg et al., Cell, 2026 (record at Google Research) — Paper abstract; neuron and cell-type counts.accessed on September 21, 2026
5. [R5]A connectomics milestone: mapping the complete male fruit fly brain. Google Research — Published 3 Sep 2026; scale, context and figures.accessed on September 21, 2026
6. [R6]Creative Commons — Attribution 4.0 International (CC BY 4.0) — Attribution, indication of changes and link to the licence.accessed on September 21, 2026
7. [R7]Connectome/GoPiGo — GitHub repository — Documented precedent of a C. elegans-based model connected to a physical robot.accessed on September 21, 2026
8. [R8]FLYNN: Robust Neural Network for Robot Navigation using Fly Brain Topology. arXiv preprint, 2026 — Female FAFB v783 connectome topology and navigation in MuJoCo. Not to be confused with MaleCNS or with a physical demonstration.accessed on September 21, 2026
9. [R12]FlyWire — Reference for scientific presentation and visual exploration of connectomes; distinct from the chosen dataset.accessed on September 21, 2026

Explore the connectome and the viewer's metadata