Droffel Isaac Brain

A virtual fruit-fly nervous system playing The Binding of Isaac, with a live 3D view of neural activity.

Droffel connects Isaac's room state to an approximate spiking model built from the MaleCNS v1.0 connectome. You can watch the controller explore, fight, collect items and respond to sensory stimulation while the same running model lights up in the brain viewer.

• Runs 165,122 neurons and 25,563,096 directed, aggregated connections.
• Shows measured cell-body positions for 140,024 neurons and selected real neuron arbors in 3D.
• Controls movement and tears through a local Isaac mod; no screen scraping or global keyboard injection.
• Avoids rocks, pits, spikes and fire; breaks poop blocking a firing lane and extinguishes tear-destructible fire.
• Fires along predicted tear trajectories while retreating and seeks pressure plates in locked, empty normal rooms.
• Collects free pickups, buys affordable shop items, and can use a charged active item or a held card during combat.
• Starts another attempt after a game-over while control remains enabled.
• Offers adaptive tactics or a no-learning mode that never reads or writes accumulated experience.

You need Python 3.11 or newer and a mod-capable installation of Isaac. The current integration is tested with Repentance+; older editions are not verified.

py -3.11 -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txt
.venv\Scripts\python.exe -m pip install -r requirements-gpu.txt
.venv\Scripts\python.exe tools\prepare_assets.py
Copy-Item game_paths.example.json game_paths.json

Edit game_paths.json to point to the folder containing isaac-ng.exe. Asset preparation downloads the pinned Godot runtime and public MaleCNS data, then builds the local graph and viewer assets. Allow several GB of free disk space. CUDA acceleration is optional; CPU fallback is available, but the full model may run below real time.

Start a run, then press F6 to enable fly control. F7 stops it. Esc pauses the game and stops control. The launcher installs only Droffel's own mod folder. Restart Isaac through the launcher after updating the mod.

The launcher sets PauseOnFocusLost=0 in Repentance+'s options.ini, preserving a backup when it changes the setting. Restart an already running game once to apply it. This lets the game continue when another window is active; Esc still pauses and stops control.

The dashboard opens a separate 3D brain window. Drag to rotate, use the wheel to zoom, and try the stimulus buttons to inspect sensory responses. The original flight-terrain prototype is not part of this release.

This is a hybrid controller, not a trained biological fly that understands a video game. Room coordinates come from the mod. Navigation, targeting, item decisions and sensory mappings are engineered. Neural firing rates gate movement and shooting, and recurrent escape-circuit activity alters evasion. Synaptic weights stay fixed; the optional learning layer adapts tactical choices and spatial risk costs. It does not train the whole connectome.

The model uses simplified leaky integrate-and-fire neurons and assumptions about neurotransmitter signs. The 3D display shows recorded anatomy where available; activity is simulated, not measured from a living animal. Item use is a general combat heuristic rather than complete knowledge of every Isaac item, card or synergy.

.venv\Scripts\python.exe -m pip install -r requirements-dev.txt
.venv\Scripts\python.exe -m pytest tests -q

The data-free tests cover controller behaviour, local bridge safeguards, Lua input handling, learning isolation and neural-kernel direction. Generated connectomes, recordings, logs, saved experience, local paths and connection tokens are excluded from git.

See third-party sources and attribution. MaleCNS data is prepared from the public FlyEM release; the simulator is derived from virtual-fly-lab. Isaac integration follows the Repentance Lua API.