News · 2026-10-18

Open Fly connects a brain simulation to a game without training it

Open Fly connects a simulated fruit-fly neural circuit to the strategy game Open Doctrines through a hand-designed mapping between game events and neural inputs and outputs. The project uses a published brain model with 138,639 neurons and roughly 15 million synaptic connections. Its repository explicitly says the system did not learn to play, making this a controller demonstration rather than evidence that a digital fly acquired a new skill.

Key facts

• The project uses the Shiu and colleagues 2024 leaky integrate-and-fire model based on the FlyWire adult-fly connectome.
• Each game turn runs 200 milliseconds of simulated neural activity before outputs are translated into orders.
• The project appeared in the supplied community feed through an October 16 post.
• The primary source is Pr1nted’s Open Fly repository and README.

A real wiring diagram meets an invented interface

A connectome describes which neurons connect to which others. It is an important biological map, but turning that map into a running simulation requires choices about how neurons respond, how connections act, and how the outside world reaches the circuit.

Open Fly uses a leaky integrate-and-fire model. In this simplified neuron model, incoming signals build up an internal quantity while that quantity also decays. Crossing a threshold produces a spike, after which the neuron’s state is reset according to the model’s rules. Connected neurons then receive the consequences of that event.

The result is a simulation grounded in anatomical connectivity, not a complete reconstruction of every biological process in a fly. The distinction becomes especially important when the circuit is given a task outside the animal’s natural setting.

For Open Fly, the project author maps events in Open Doctrines to sensory stimuli and maps activity from descending neurons to game orders. Those choices are part of the controller. They determine what the simulated circuit can perceive and how its activity affects the game.

Why the mapping matters as much as the brain label

Think of connecting a weather sensor to a musical instrument. A programmer can decide that rising temperature changes pitch and falling pressure increases volume. The sensor genuinely drives the music, but it has not learned composition, and the mapping strongly shapes what the listener hears.

Open Fly’s interface plays a similar causal role. Game events become selected neural inputs; selected neural outputs become strategic actions. Observing interesting behavior therefore does not isolate the contribution of the biological wiring from the contribution of the interface design.

The README makes the limitation unusually explicit: “nothing about it learned to play.” That statement rules out the most tempting interpretation of a biological-brain headline. There is no demonstrated training process in which the simulated fly gradually discovers a strategy through experience.

Our lesson on learning by imitation explains another route to building a controller. Neither a hand-designed interface nor the use of an anatomical network should be casually described as that kind of learning.

What the project’s checks establish

The repository says its browser implementation matches the original Brian2 simulation spike-for-spike on scripted inputs. That is a useful implementation check: under the tested conditions, the software produces the expected event sequence.

It is not the same as showing that a biological fly would behave similarly in the game, that the mapping captures natural sensory processing, or that the resulting orders represent an acquired skill. Those would require different experiments and comparisons.

The publicly inspectable implementation is the reason the project merits attention. Readers can examine the chosen input and output mappings and ask what happens under different conditions. Sensible comparisons would include altered mappings or alternative circuits. Those are suggestions for evaluating the demonstration, not experiments the dossier establishes were completed.

Keep the Odyssey story separate

A different claim circulating on October 18 described a fly-brain simulation navigating environments generated by Odyssey-3. The dossier did not verify a technical report, code, or protocol for that integration. Odyssey’s technical introduction and October 8 launch page describe the world model and other policy demonstrations, but do not document the claimed fly connection.

Open Fly’s source supports a strategy-game controller. It does not authenticate the separate generated-world demo. Combining the two would turn a documented project and an unverified integration claim into a much stronger scientific story than either source supports.

Ground Truth has also covered fruit-fly connectome research. That wider field provides context for why anatomical maps are exciting, without making this hobby project a new neuroscience result.

An interesting controller is enough

Open Fly demonstrates an inspectable way to put a connectome-based dynamical system into an artificial task. Its value lies in exposing the mechanics of that experiment and making the interface available for scrutiny.

The honest limit is equally clear: a simulated circuit producing game commands does not establish learning, biological fidelity, or a complete digital animal. The most useful next question is which parts of the observed behavior depend on the circuit and which depend on the human-designed translation around it.

Primary source, verified: read the paper →

Ground Truth. (2026, October 18). Open Fly connects a brain simulation to a game without training it. Ground Truth. https://groundtruth.day/news/open-fly-connectome-controller-did-not-learn.html

@misc{groundtruth:open-fly-connectome-controller-did-not-learn,
  title  = {Open Fly connects a brain simulation to a game without training it},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/open-fly-connectome-controller-did-not-learn.html}
}

Topics: neuroscience · simulation · open-source · spiking-neural-networks