https://github.com/amanuel-1/autonomous-drone

A 3D autonomous drone simulation with AI-powered flight capabilities using deep reinforcement learning. Features realistic physics, LiDAR obstacle detection, and a neural network that learns to navigate complex environments through both reinforcement learning and imitation learning from human demonstrations.
https://github.com/amanuel-1/autonomous-drone

autonomous-drone deep-learning imitation-learning lidar-sensors reinforcement-learning threejs

Last synced: 4 months ago
JSON representation

A 3D autonomous drone simulation with AI-powered flight capabilities using deep reinforcement learning. Features realistic physics, LiDAR obstacle detection, and a neural network that learns to navigate complex environments through both reinforcement learning and imitation learning from human demonstrations.

• Host: GitHub
• URL: https://github.com/amanuel-1/autonomous-drone
• Owner: Amanuel-1
• License: other
• Created: 2025-06-08T18:26:18.000Z (4 months ago)
• Default Branch: main
• Last Pushed: 2025-06-09T09:17:29.000Z (4 months ago)
• Last Synced: 2025-06-09T10:25:26.414Z (4 months ago)
• Topics: autonomous-drone, deep-learning, imitation-learning, lidar-sensors, reinforcement-learning, threejs
• Language: TypeScript
• Homepage:
• Size: 20.9 MB
• Stars: 1
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Autonomous Drone Simulation



A sophisticated 3D drone simulation built with Next.js, Three.js, and React Three Fiber, featuring AI-powered autonomous flight capabilities using deep reinforcement learning.

![Drone Simulation Demo](public/droneSim.gif)

## 🚁 Features

### Core Simulation

- **Realistic 3D Drone Physics**: Full 6-DOF movement with realistic tilt, thrust, and inertia

- **Advanced Flight Controls**: Manual control with keyboard inputs for takeoff, landing, movement, and camera control

- **Dynamic Camera System**: Movable gimbal camera with tilt and rotation controls

- **Comprehensive Environment**: Buildings, skyscrapers, dense forests, and training obstacles

### AI & Machine Learning

- **Deep Reinforcement Learning**: Neural network-based autonomous flight using Q-learning

- **Imitation Learning**: Record and learn from human demonstrations

- **Advanced Reward System**: Shaped rewards for collision avoidance, mission completion, and flight efficiency

- **Real-time Training**: Live training with exploration/exploitation balance

### LiDAR & Sensing

- **Optimized LiDAR System**: 16 spherical rays providing 360° 3D obstacle detection

- **Real-time Visualization**: Visible ray casting with intersection markers

- **Performance Optimized**: Reduced from 52 to 16 rays for better simulation performance

### Mission System

- **Dynamic Mission Generation**: Randomized start and target positions

- **Landing Challenges**: Precision landing requirements within target zones

- **Progress Tracking**: Real-time distance and completion monitoring

### Training Environment

- **Configurable Difficulty**: Beginner, intermediate, and advanced training modes

- **Dynamic Obstacles**: Moving platforms, pendulums, and wind zones

- **Comprehensive Scenarios**: Gates, tunnels, narrow passages, and maze walls

## 🛠 Technology Stack

- **Frontend**: Next.js 15, React 19, TypeScript

- **3D Graphics**: Three.js, React Three Fiber

- **Styling**: Tailwind CSS 4

- **Icons**: Lucide React

- **AI/ML**: Custom neural network implementation with reinforcement learning

- **Package Manager**: pnpm



## 🚀 Getting Started

### Prerequisites

- Node.js 18+

- pnpm (recommended) or npm/yarn

### Installation

1. Clone the repository:

```bash

git clone 

cd autonomous-drone

```

2. Install dependencies:

```bash

pnpm install

```

3. Start the development server:

```bash

pnpm dev

```

4. Open [http://localhost:3000](http://localhost:3000) in your browser

## 🎮 Controls

### Manual Flight Controls

- **Movement**: Arrow keys (↑↓←→)

- **Altitude**: Shift + ↑↓

- **Rotation**: Shift + ←→

- **Takeoff**: T

- **Landing**: L

- **Hover**: H

- **Camera**: I/K (tilt), J/O (rotate)

### AI Controls

- **Toggle AI Mode**: Switch between manual and autonomous control

- **Training**: Enable/disable real-time learning

- **Recording**: Record demonstrations for imitation learning

- **Save/Load**: Export and import trained models

## 🧠 AI Architecture

### Neural Network

- **Input Size**: 40 features

  - Position, velocity, rotation (9)

  - Drone status (4)

  - LiDAR readings (16)

  - LiDAR indicators (3)

  - Flight status (2)

  - Mission info (6)

- **Architecture**: 256→128→64 hidden layers

- **Output**: 9 possible actions (movement + hover)

### Training Features

- **Reinforcement Learning**: Q-learning with experience replay

- **Imitation Learning**: Learn from human demonstrations

- **Reward Shaping**: Complex reward system for optimal behavior

- **Auto-respawn**: Continuous training with automatic episode management

## 📊 Performance Optimizations

### LiDAR System

- **Reduced Ray Count**: Optimized from 52 to 16 spherical rays

- **68% Performance Improvement**: Significant reduction in computational overhead

- **Maintained Coverage**: Full 3D spatial awareness with spherical distribution

### Neural Network

- **Compact Architecture**: Reduced input size from 77 to 40 features

- **Efficient Training**: Smaller network for faster convergence

- **Real-time Inference**: Optimized for live decision making

## 🎯 Mission Types

- **Basic Navigation**: Fly from start to target position

- **Precision Landing**: Land within 3-meter target zones

- **Obstacle Avoidance**: Navigate through complex environments

- **Altitude Challenges**: Maintain optimal flight altitudes

- **Speed Optimization**: Complete missions efficiently

## 🔧 Configuration

### Training Parameters

- **Episode Length**: 3000 steps maximum

- **Learning Rate**: 0.0005

- **Exploration**: Epsilon-greedy with decay

- **Replay Buffer**: 100,000 experiences

- **Batch Size**: 32 samples

### Environment Settings

- **World Size**: 200m × 200m

- **LiDAR Range**: 25m maximum

- **Altitude Limits**: 0-100m

- **Mission Distance**: 15-60m

## 📈 Monitoring & Analytics

- **Real-time Metrics**: Episode rewards, collision counts, success rates

- **Training Progress**: Loss curves, exploration rates, performance trends

- **Flight Data**: Position tracking, LiDAR readings, action history

- **Model Management**: Save/load trained networks and demonstrations

## 🤝 Contributing

1. Fork the repository

2. Create a feature branch (`git checkout -b feature/amazing-feature`)

3. Commit your changes (`git commit -m 'Add amazing feature'`)

4. Push to the branch (`git push origin feature/amazing-feature`)

5. Open a Pull Request

## 👨‍💻 Author

**Amanuel Garomsa**

- 🎓 Computer Science Graduate

- 🏢 Currently working at Icoglabs, SingularityNet

- 📧 Email: amanuelgaromsa@gmail.com

- 🔬 Personal Research Project

## 📝 License

MIT License © 2025 Amanuel Garomsa

This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.

This is a personal research project exploring autonomous drone flight using deep reinforcement learning and imitation learning techniques.