Multi-Cellular Demo

A 2D simulation of evolving multi-cellular agents. Particles move on a grid, consume food from cells, form bonds, reproduce (consuming food at the midpoint), and mutate. Genome types include neural networks and a simpler gene-based genome.

Build and run

Requires .NET 9.0.

cd Demos/MultiCellularDemo
dotnet run

Optional arguments:

dotnet run -- --seed 12345              # Somwhat reproducable run with seed 12345
dotnet run -- --genome neural           # Use neural genome (see Genome types)
dotnet run 12345                        # Seed only (bare integer)
dotnet run -- --seed 42 --genome deep   # Combine options

Command-line arguments

Argument	Description
--seed <number>	RNG seed for the run. Use the same seed to reproduce a run.
--genome <type>	Genome type for the initial population (see below).
<number> (bare)	If a single integer is passed, it is used as the seed.

Genome types

Pass --genome <type> with one of:

• neural – Feedforward neural net (71 inputs, 1 hidden layer, 36 outputs). Includes velocity, age, bonds, nearby cells, food, and outputs e.g. force, hue, bond strength, bind/unbind/reproduce targets, mutation rate, nearby attraction, nearby food attraction.
• deep / deepneural – Deeper neural net (71 inputs, 3 hidden layers, 36 outputs). Same I/O as neural.
• chemistry – Reduced-input neural net (38 inputs, no velocity/age, distance-only for bonds/nearby). 35 outputs (no nearby food attraction).
• particlegene / gene – (default) Simple gene-based genome, not a neural net.

If --genome is omitted or unrecognized, particlegene is used.

---

UI

• Window – 1280×720, 60 FPS target. Black background; world is drawn with a 2D camera.

Camera and view

• Pan – Click and drag to move the camera.
• Zoom – Mouse wheel: scroll up to zoom in, down to zoom out. Zoom is clamped between 0.01 and 10.
• Particle outline – White outline on particles is drawn only when zoomed in (zoom ≥ 0.4). When zoomed out, only filled colors are shown so hues remain visible.

Selection

• Select particle – When paused, click in the world to select a particle under the cursor. When the selected particle dies, selection can switch to a bond partner or a particle near the camera target.
• Camera follow – When a particle is selected, the camera smoothly follows it.
• Selected panel – Left side shows: position, hue, bond count, genome gene values, and last neural output (force, hue, bond strength, bind/unbind/reproduce, mutation rate, nearby attraction, etc.).
• Network view – When the selected particle uses a neural genome, a small network diagram appears in the bottom-right (layers, weights, activations).

HUD (left side)

• Population – Current particle count.
• Generation – Median generation of particles.
• Births/s – Births per second over a 5-second window.
• Total food – Sum of food in all grid cells (explored + unexplored at initial value).
• Max food/s – Maximum food regeneration per second (based on map size and food params).
• Hue histogram – Bar chart of particle count per hue bin (36 bins).
• Births chart – Time series of births/s (white line). Green line = max food/s reference.

Top-right

• Seed: <number> – Current run seed. Click to copy to clipboard (text highlights for a few seconds).
• FPS – Current frames per second.

Keyboard

• F5 – Pause / resume simulation. When paused, you can click to select a particle.

---

Where parameters are defined

Tunable constants are spread across a few files. Change them in code and rebuild.

Context (Context.cs)

Food grid and global simulation state:

• FoodCellSize – World units per grid cell (500). Cell centers are used for food and for “nearest food” attraction.
• MaxFood – Cap per cell (100). Food regenerates toward this.
• InitialFoodPerCell – Amount when a cell is first created (0).
• SecondsPerFood – Regeneration rate: each cell gains 1/SecondsPerFood per second (e.g. 5 ⇒ 0.2/s per cell).

Simulation (Simulation.cs)

Map and population:

• MapSize – World size in both X and Y (10000).
• MinCameraSelectionBorderDistance – When selecting a particle for camera follow, only consider particles at least this far from the map edge (500).
• MaxBirthsPerSecond – Birth rate budget (400). Budget accumulates each frame and is spent per birth.
• PopulationCap – Max particles (1200). Excess is culled (oldest first) when over cap.

GridCellSize is taken from Context.FoodCellSize.

Particle (Particle.cs)

Physics, bonds, and rendering:

• Radius – Particle radius for drawing and hit-test (5).
• NearbyRadius – Max distance to include another particle in “nearby” input (1000).
• MaxBondDistance – Bonds break if partners are farther than this (200).
• ForceScale – Scale for force outputs (500). Output 0.5 = neutral; (output − 0.5)×2×ForceScale gives force.
• SpringRestLengthMin / SpringRestLengthMax – Rest length for bonds from SpringDistance output (10–100).
• BondAgeNormScale – Bond age normalized by this for inputs (10).
• MinBondReproductionTime – Minimum bond age before reproduction is allowed (0.5 s).
• MaxBondingPartnersMin / MaxBondingPartnersMax – Range for decoding MaxBondingPartners output (1–10).
• MinMutationRate / MaxMutationRate – Range for mutation rate from genome output (0.01–0.5).
• MaxVelocityCap – Velocity clamped to this magnitude (1600).
• AgeNormScale – Age in seconds divided by this for AgeNorm input (60).
• MinZoomForOutline – White outline on particles only when zoom ≥ this (0.4).
• HueTickRate – Interval (s) between applying genome hue output to the particle (0.1).

Program (Program.cs)

UI and run setup:

• BirthsAvgWindowSeconds – Window for births/s average (5).
• BirthsChartMaxSamples – Max points in the births chart (300).
• SeedHighlightDuration – How long the seed text stays highlighted after copy (3 s).
• HueBins – Number of hue histogram bars (36).
• cameraSmoothSpeed – Camera follow smoothing (4).
• Chart sizes, network panel size, etc. are in local constants in the draw loop.

Genomes

• NeuralGenome – NeuralGenome.cs: InputDim, HiddenDim, OutputDim, MaxBondsEncoded, MaxNearbyEncoded, mutationStrength in CloneMutate.
• DeepNeuralGenome – DeepNeuralGenome.cs: LayerDims, OutputDim, same mutation/encoding constants.
• ChemistryGenome – ChemistryGenome.cs: InputDim, OutputDim (35, no nearby food attraction), LayerDims, etc.
• ParticleGene – ParticleGene.cs: gene count and mutation parameters.

---

Summary

Run with dotnet run; use --seed for reproducibility and --genome to choose neural, deep, chemistry, or particlegene. Pan and zoom with mouse; pause with F5 and click to select a particle and inspect its genome and last output. Tweak behavior by editing the constants in Context.cs, Simulation.cs, Particle.cs, and Program.cs as listed above.