🚁 Genesis Drone Env

High-Fidelity Drone Simulation Environment based on Genesis

Documentation | Genesis Engine

---

Features • Installation • Demos • FPV Hardware • Citation

📖 Introduction

Genesis Drone Env provides a robust playground for drone research, ranging from Reinforcement Learning (RL) to classical Geometric Control. Included in the official Genesis ecosystem, this repository serves as a foundation for developing complex aerial robotics algorithms.

🔖 Related Work (Our Work)

1. FLARE: Agile Flights for Quadrotor Cable-Suspended Payload Systems via Reinforcement Learning (Accepted by IEEE RA-L) (Github Code)

✨ Features

• 🚀 Reinforcement Learning: Ready-to-use environments for training tracking policies (PPO included).
• 📐 Geometric Control: Concise implementation of SO(3)/SE(3) controllers for precise trajectory tracking.
• 🎮 Hardware-in-the-Loop (HIL): Connect your real RC transmitter via Flight Controller (FCU) to fly FPV in the simulator.
• ⚙️ Highly Configurable: Easy tuning of flight parameters, physics settings, and reward functions via YAML.

---

💻 Installation

It is recommended to use a virtual environment (conda) to manage dependencies.

# 1. Create environment
conda create -n genesis_drone python=3.11 # Requires Python >= 3.10
conda activate genesis_drone

# 2. Install Genesis (Ensure you have the latest version)
# Visit https://github.com/Genesis-Embodied-AI/Genesis for detailed instructions

# 3. Clone and install this repository
git clone https://github.com/KafuuChikai/GenesisDroneEnv.git
cd GenesisDroneEnv
pip install -e .

---

🎬 Demos & Usage

1. RL Tracking Task

Train or evaluate a policy to track a moving target.

Evaluate Pretrained Model:

python scripts/eval/track_eval.py

Train Your Own Policy:

python scripts/train/track_train.py 

Note: Training typically converges around the 200th step. You can fine-tune the reward scale in rl_env.yaml.

2. SE(3) Geometric Controller

Replace the neural network with a classical geometric controller for precise maneuvers.

Run Trajectory Tracking:

python scripts/eval/se3_controller_eval.py --use-trajectory

Tips:

• --use-trajectory: Enables circle trajectory tracking.
• Without the flag, it defaults to Waypoint Mode.

The controller implements the framework proposed in:

• Geometric tracking control of a quadrotor UAV on SE(3)
• Control of Quadrotors Using the Hopf Fibration on SO(3)

3. 🎮 Hardware-in-the-Loop (FPV)

Fly the simulated drone using your real Radio Controller (RC) via a Flight Controller (FCU) bridge.

Hardware Setup

1. Prepare FCU: Use an STM32H743 FCU.
2. Flash Firmware: Flash the custom HEX file: betaflight_4.4.0_STM32H743_forRC
3. Connect:
   • Power the FCU via USB-C.
   • Connect the FCU's UART port to your PC using a USB-to-TTL module.

Software Configuration

1. Check your serial port ID (e.g., /dev/ttyUSB0) using:

ls /dev/tty*

2. Update the USB_path parameter in flight.yaml (or relevant config) to match your port.

Run FPV Mode

python scripts/eval/rc_FPV_eval.py

🛠 Configuration

Customize the simulation to fit your needs by editing the YAML files in the config/ directory:

Config Type	File Path	Description
Flight Dynamics	config/*/flight.yaml	PID gains, physical properties (mass, inertia).
Environment	config/*/genesis_env.yaml	Physics engine settings, rendering, scene setup.
RL Hyperparams	config/*/rl_env.yaml	Reward functions, observation space, training steps.

---

🤝 Acknowledgement

This repository is inspired by the following ground-breaking work:

Champion-level drone racing using deep reinforcement learning (Nature 2023)

We acknowledge the contributions of the open-source community that make this project possible.