A single-file, browser-based evolution sandbox. Agents (prey & predators) use small neural networks with mutable topology (nodes/layers/edges) and weights to forage, flee, hunt (including predator-vs-predator), and reproduce. Traits like color, attack strength, and network structure evolve over time. Includes a live NN viewer, follow modes, and save/load for brains.
- Variable-topology brains: hidden layers & node counts evolve; connections add/remove.
- Sensing: food aggregate, kin aggregate, and top-K (4) nearest agents with angle, closeness, type (prey/pred), and kin flags (parent/child/sibling).
- Behavior: predators hunt prey & other predators; kin-penalties discourage eating parent/child/sibling.
- Traits: color (HSL) is heritable + mutatable; predators have a damage output node.
- Viewer: curved-edge NN graph with live activations + labeled outputs; per-agent insights (gen, fitness, I/O snapshot).
- Camera/Follow: many follow modes (best, latest gen, most nodes/layers/conns, density, fittest, fastest, etc).
- Performance: timescale slider with sub-stepping; optional trail/vision rendering.
- Download/clone and open
evolution_simulator.htmlin a modern browser. - Use the sidebar to tune populations, energy, evolution rates, perception/motion, and timescale.
- Click an agent to inspect its brain; change Follow to auto-track interesting individuals.
- Run/Pause, Step, Reset
- Follow selector (many modes)
- Save brain (selected) • Load → Selected • Default Prey/Predator (reseeds on Reset)
- Optional Trails & Show vision toggles
Space= Pause/RunShift + Space= Step one tick
This project is a compact playground for emergent behavior and neuro-evolution with minimal dependencies. It’s great for experimenting with sensing abstractions, selection pressure, and genome/topology mutations to see what strategies emerge.