react-biosim
An environment for evolutionary simulations in the browser, inspired by David R. Miller's video "I programmed some creatures. They Evolved.".
Creatures live on a grid and are controlled by small neural networks built from their genes. At the end of each generation, the creatures that meet a goal survive and pass their genes, with occasional mutations, to the next generation. After a few generations, behaviour that looked random becomes purposeful.
Live demo: https://biosim.rdalmau.com
Quick start
- Open the Scenario tab and pick a scenario in Demo scenarios.
- Watch the fitness curve rise in the Simulation tab (Survival rate (%)).
- Click a species in the Population tab to see its brain.
- Change the rules or draw a new map in the Scenario tab, then click Restart.
Use the control bar at the top to pause the simulation or change its speed. Set Immediate steps to 200 to run at full speed.
Included scenarios
| Scenario | What to watch |
|---|
| davidrmiller examples 1, 2 and 4 | The demos from the video, some resumed at a late generation |
| Vertical boxes | Creatures learn to reach the right box |
| Turn right / Turn right generation 5574 | Creatures learn to reach the bottom-left corner of the map |
| Carlos' original | Creatures learn to stay in the centre |
| NW7 generation 13780 | A long run resumed at generation 13780 |
How the simulation works
- World. A square grid (100 × 100 cells by default) with obstacles and areas: spawn, reproduction and health.
- Creatures. Each creature fills one cell. Its genome is a list of 32-bit genes, and each gene is one weighted connection in its brain.
- Brain. Sensors (position, age, nearby obstacles, prey…) feed a small neural network that drives actions (move, reproduce, attack…).
- Generations. A generation lasts a fixed number of steps. When it ends, a selection method picks the survivors, for example those standing inside a reproduction area.
- Inheritance. A population strategy fills the next generation with copies of the survivors' genomes. A copy can mutate: a gene changes, or a gene is added or removed.
- Metabolism and genera. Optional mass and water, and genera (plants, herbivores, carnivores) for predator–prey experiments.
Details: sensors and actions, population strategies and selection methods and all parameters.
App tabs
| Tab | Use it to |
|---|
| Scenario | Load a demo scenario, a .sim file or a saved scenario; change world, generations, neural networks, mutations, sensors, actions and the map, then apply with Update simulation (keeps the creatures) or Restart |
| Simulation | Simulation code, seed and display options; plot fitness per generation and check the water |
| Population | See the top species, a creature's genome and its neural network, live or as a diagram |
| Export | Save your own scenarios, save the simulation to a .sim file or as JSON, export images and GIFs |
| Tools | Record the simulation events with the logger |
| About | Credits and links |
Documentation
All pages are listed in docs/.
Development
Requirements: Node.js 20.9 or later (required by Next.js 16).
npm install
npm run dev # development server at http://localhost:3000
npm test # Jest unit tests (__tests__/)
npm run build # production build
npm run lint
Project layout:
| Folder | Contents |
|---|
app/ | Next.js entry page and layout |
components/ | React UI: tabs, control bar, canvas and shared inputs |
components/simulation/store/ | Jotai atoms that connect the UI to the simulation |
simulation/ | The simulation engine, written in plain TypeScript with no React |
hooks/ | React hooks |
__tests__/ | Unit tests for the engine |
public/ | Demo scenario files loaded by the Scenario tab |
Built with TypeScript, React, Next.js, Jotai, Tailwind CSS, Chart.js and D3.
Credits
- Original model and C++ simulator: David R. Miller, biosim4.
- TypeScript/React port and original UI: Carlos Peña, react-biosim.
- This fork, by taganz, adds scenarios, a map editor, alternative selection and population methods, metabolism and genera, logging, GIF export and more.
This is a hobby project under development. See Known issues and to-do.
License: MIT, see LICENSE.
