A 25-year-old software worker loses his job to AI. He inherits an old house from his grandfather — and discovers an unfinished robot inside.
A narrative-driven training simulator about trust, purpose, and the machines we build.
How To Train Your AI is a Unity 6 first-person experience where you teach a broken robot to think, move, and solve problems — one mini-game at a time. The irony is the point: the same technology that displaced the player becomes the thing that restores his purpose.
Through three distinct training modules, the robot evolves from an unreliable prototype into a capable companion. Every mistake you make has consequences. Every improvement feels earned.
You are Nour — a software engineer who just got laid off to AI automation. While packing up your life, a lawyer hands you a letter: your grandfather left you his old house.
Inside, buried under dust and memories, you find a half-finished robot and a series of recorded messages. Your grandfather was building something — something he never got to finish. His final message is corrupted, but the last thing it says is clear:
"Trust the robot. It will show you what matters."
The journey begins in the living room, but it ends somewhere you didn't expect.
The game is built around three training mini-games, each teaching the robot a fundamental skill. Performance is measured, scored, and — critically — it matters.
The robot's movement system is unstable. You need to calibrate it.
| Challenge | What You Do |
|---|---|
| Drift Handling | The robot drifts left or right. Counter-steer to stay on course. |
| Camera Alignment | The camera pitch is offset. Return it to the target angle. |
| Speed Consistency | Speed wobbles unpredictably. Maintain a steady pace. |
Your scores determine how reliable the robot is going forward. Nail it, and the robot barely stumbles. Fail, and you'll be fighting random faults for the rest of the game.
Scoring: Drift (40%) + Camera (25%) + Speed (35%) → Normalized 0–100 scale
| Tier | Score | Result |
|---|---|---|
| Excellent | 90+ | Minimal faults |
| Good | 70–89 | Occasional faults |
| Average | 50–69 | Frequent faults |
| Fail | < 50 | Retry required |
A top-down grid challenge. The robot needs to collect an audio card while managing a limited energy budget.
- Click to move the robot across tiles
- Different tiles cost different energy — some are efficient, some are expensive
- Energy depletion = mission failure
- Path efficiency is measured against the ideal shortest path
This teaches the robot to make smart decisions under constraints — not just fast ones.
Scoring: Energy Efficiency (40%) + Path Efficiency (35%) + Decision Quality (25%)
A sci-fi lab. Push devices to their correct positions. Sounds simple — until the puzzles get layered.
| Task | Mechanic |
|---|---|
| Exact Placement | Push specific devices to their exact matching target tiles |
| Group Placement | Group devices by type — any device of the correct group works |
| Size Ordering | Sort devices by size rank in the correct sequence |
- Right-click to pathfind to a destination
- Press E to push objects one grid step
- Objects block movement and future devices can't be pushed
- Deadlock detection resets the task if you get stuck
No timer. No score pressure. The challenge is pure spatial logic.
After completing Mini-Game 1, the robot starts experiencing random fault events based on your training quality:
| Fault | Effect |
|---|---|
| Drift Fault | Temporary yaw drift during movement |
| Camera Fault | Temporary pitch offset in the robot's camera |
| Speed Fault | Sprint gets blocked or canceled |
Better training = fewer interruptions. The robot's reliability is a direct reflection of your effort.
- Player/Robot Control Switching — Tab to toggle between controlling the player and the robot, with Cinemachine camera blending
- Persistent Robot Stats — Your training performance carries forward through the entire game via ScriptableObject data
- Event-Driven Architecture — All systems communicate through C# events, keeping modules decoupled
- A* Pathfinding — Custom grid-based pathfinding with energy costs, diagonal support, and obstacle detection
- Deadlock Detection — MG3 automatically detects when puzzles are unsolvable and resets
- Scene Transitions — Smooth fade-to-black transitions between story and gameplay segments
- Narrative Delivery — Corrupted grandfather messages that slowly reveal the story
- Developer Tools — Built-in skip shortcuts (F8 in MG2, Enter in MG3) for testing
| Action | Key |
|---|---|
| Move | WASD |
| Look Around | Mouse |
| Interact | E |
| Toggle Robot Control | Tab |
| Sprint | Left Shift |
| Push Object | E (when adjacent) |
| Pathfind (MG2/MG3) | Right-Click |
| Move to Tile (MG2) | Left-Click on tile |
HowToTrainYourAI/
├── Assets/
│ ├── Scripts/
│ │ ├── Managers/ # ControlManager, SceneTransitionFader
│ │ ├── Robot/ # RobotMovement, CameraLook, FollowPlayer
│ │ ├── Mini Game 1/ # Challenges, Scoring, Fault System
│ │ ├── Mini Game 2/ # GridManager, EnergySystem, TileClickMover
│ │ ├── Mini Game 3/ # PushPuzzle, Pathfinder, TaskValidation
│ │ └── UI/ # Navigation, SlideUI, TimerUI
│ ├── Scenes/
│ │ ├── Nour/ # Main Scene (house, player, robot)
│ │ ├── Omar/ # Mini-Game 1
│ │ ├── Oraby/ # Mini-Game 2
│ │ ├── Aya/ # Mini-Game 3, Post-Credits
│ │ └── Dialogue/ # Message & Office dialogue
│ ├── Data/
│ │ ├── Robot/ # RobotStats_Main.asset
│ │ └── MiniGames/ # Learning profile ScriptableObjects
│ ├── Prefabs/ # Player, Robot, Home, UI
│ └── Mini Game 3/ # Lab GLB, animations
├── Base Rules/ # Design documentation
└── Documentation/ # Technical specs and reports
| Pattern | Where |
|---|---|
| Singleton | SceneTransitionFader, MG1InstructionSequenceController |
| ScriptableObject | RobotStatsSO, MiniGame1LearningProfileSO, MiniGame2LearningProfileSO |
| State Machine | Phase enums in MG1, MG2, MG3 |
| Strategy / Template | MiniGame1ChallengeBase → Drift, Camera, Speed challenges |
| Observer | UI scripts subscribe to manager events |
| Flow Coordinator | MG1ToMG2FlowCoordinator for story transitions |
| Component | Grid tiles, pushable devices, target slots as composable MonoBehaviours |
| Technology | Version | Purpose |
|---|---|---|
| Unity | 6000.5.1f1 | Game engine |
| Universal Render Pipeline | 17.5.0 | Rendering |
| Cinemachine | 3.1.6 | Camera system |
| Input System | 1.19.0 | Player input |
| AI Navigation | 2.0.13 | NavMesh |
| glTFast | 6.19.0 | 3D model import |
| TextMeshPro | 2.5.0 | UI text |
| Timeline | 1.8.12 | Cutscenes |
- Unity Hub
- Unity 6000.5.1f1 (install via Unity Hub)
git clone https://github.com/NourEldeenMahmoud/HowToTrainYourAI-Game.git
cd HowToTrainYourAI-Game
git lfs install
git lfs pullOpen the folder in Unity Hub. First open will regenerate the Library/ folder locally.
Open the Main Scene to start.
| Name | Role | GitHub |
|---|---|---|
| Nour | Core Systems, MG1, Robot Logic | @NourEldeenMahmoud |
| Omar | Mini-Game 1 | @OmarAbouelkheirr |
| Oraby | Mini-Game 2 | @abdalrhman541 |
| Aya | Mini-Game 3, Post-Credits | @AyaSheta13 |
This project is for educational purposes. See the repository for license details.