中文 | English

XPlane Gym: A Reinforcement Learning Environment Compatible with X-Plane

XPlaneGymEnvs is an X-Plane flight simulator environment compliant with the OpenAI Gym interface, specifically designed for reinforcement learning research. It provides seamless integration with the X-Plane simulator, supports both discrete and continuous action spaces, and can be used to train agents to perform flight control tasks.

Installation Requirements

• X-Plane 12 Flight Simulator（Perhaps it will also work in lower versions）
• Python 3.8+
• gymnasium
• numpy

Installation

# Clone the repository
git clone https://github.com/Picaun/XPlaneGymEnvs.git
cd XPlaneGymEnvs

# Install the project and its dependencies
pip install -e .

Available Environments

• XPlane-v0: Basic environment, configurable as discrete or continuous action space
• XPlane-custom-v0: custom environment, Import XPlaneEnv class to personalize the environment

Quick Start

1. Installation

pip install -e .

2. Launch X-Plane

1. Start the X-Plane flight simulator
2. Ensure that the UDP communication port is set to 49000 (default value) in "Settings > Network"
3. Starting a new flight, you can set the weather conditions and flight duration by yourself (note that this function is not yet directly controlled with X-Plane 12 in the XPlaneGymEnvs interface)

3. use agent_examples

cd agent_examples/dqn_example

python train_dqn.py

Environment Parameter Configuration

Various parameters can be configured when creating the environment:

env = gym.make(
    "XPlane-v0",
    ip='127.0.0.1',                # X-Plane IP address
    port=49000,                    # X-Plane UDP port
    timeout=1.0,                   # Communication timeout
    pause_delay=0.05,              # Action execution pause delay
    starting_latitude=37.558,      # Initial latitude (default near Seoul Gimpo International Airport)
    starting_longitude=126.790,    # Initial longitude (default near Seoul Gimpo International Airport)
    starting_altitude=3000.0,      # Initial altitude
    starting_velocity=100.0,       # Initial velocity
    starting_pitch_range=10.0,     # Initial pitch angle random range
    starting_roll_range=20.0,      # Initial roll angle random range
    random_desired_state=True,     # Whether to use random target attitude
    desired_pitch_range=5.0,       # Target pitch angle random range
    desired_roll_range=10.0,       # Target roll angle random range
    action_dim=2                   # Continuous action space dimension
)

Acknowledgements

This project is based on the following open source projects:

• XPlaneConnectX
• GYM_XPLANE_ML
• Gymnasium