Tutorials
Tutorials are how most people actually learn IgnitionAI. They’re different from the How it works pages in one important way: a tutorial builds a working thing from scratch, one step at a time, with every file shown and every decision explained. You don’t need to jump around — you start at the top, follow the steps, and at the bottom you have a trained agent.
Available tutorials
| Tutorial | Difficulty | Time | Topic |
|---|---|---|---|
| GridWorld · Start here | Beginner | ~30 min | Build a custom TrainingEnv, train a DQN agent, watch it find the shortest path in a 7×7 grid. |
| CartPole: custom observations | Beginner+ | ~25 min | What the agent sees is what the agent can learn. Three obs variants, one lesson. |
| MountainCar: reward shaping | Intermediate | ~30 min | The classic sparse-reward trap — watch DQN fail, then fix it with dense shaping. |
| CartPole 3D (R3F) | Intermediate | ~40 min | Wire the Quickstart CartPole into a React Three Fiber scene with shadows and orbit controls. |
| Car Circuit | Advanced | ~60 min | The hero demo: oval track, non-trivial reward, chase cam, the lap-bonus bug you’ll write by accident. |
| Export to Unity (ONNX) | Advanced | ~60 min | The full Train → Export → Convert → Deploy pipeline with Unity Sentis. |
| Drone Navigation (physics) | Advanced | ~40 min + training | The flagship. Quadcopter with hand-rolled rigid-body physics, 6-DOF control, honest expectations about convergence, and the IgnitionAI philosophy section. |
Suggested order
The tutorials are independently runnable, but there’s a natural progression:
- GridWorld → custom env basics, zero-config training
- CartPole observations → how observation design shapes learning
- MountainCar reward shaping → how reward design shapes learning
- CartPole 3D → your first R3F integration
- Car Circuit → the advanced 3D showcase
- Drone Navigation → real physics, 6-DOF control, friction vs difficulty
- Export to Unity → the deploy story, end-to-end
If you’re in a hurry, do GridWorld → CartPole 3D → Export to Unity. That’s the minimum viable path from “never heard of IgnitionAI” to “shipped a trained policy in a game engine.”
Prerequisites for every tutorial
Before starting any tutorial, make sure you have:
- Node.js 20 or later.
- A package manager —
npm,pnpm, oryarnall work; the tutorials usenpmfor the install commands. - A bundler that handles ES modules — Vite (recommended), Next.js, or Webpack 5. The tutorials use Vite.
- Basic comfort with TypeScript. You don’t need to be an expert, but if
class Foo implements Barmakes you panic, read a TypeScript crash course first.
You do not need:
- Prior RL knowledge. The Algorithms pages cover the theory; the tutorials cover the practice.
- A GPU. Everything runs in the browser on whatever WebGPU/WebGL your machine already has.
- Python. At all. For anything. Except the one-time ONNX conversion step if you’re deploying outside the browser.
Previous: ← React Three Fiber · Next: GridWorld · Start here →
Last updated on