TUY3 / gym-dogfight3d

DogFight3d: a 3d air combat reinforcement learning environment

DogFight3d is a reinforcement learning environment for training fixed-wing aircraft 1v1 air combat, it has approximate real dynamics.

Dependencies

• gym, numpy, transform3d
• python3.8
• linux or win

Installation

gym-dogfight3d is pip installable using its GitHub:

pip install git+https://github.com/TUY3/gym-dogfight3d

or

git clone https://github.com/TUY3/gym-dogfight3d
cd gym-dogfight3d
pip install -e .

Environment

observation space

# ownship
position(x,y,z)
linear speed
linear acceleration
health level
cap(yaw),pitch,roll
thrust_level
# opponent
position(x,y,z)
linear speed
cap(yaw),pitch,roll

action space

continuous action space, including throttle,elevator,aileron,rudder

Example

env = gym.make('Dogfight3d-v0')
obs_space = env.observation_space
action_space = env.action_space
init_obs = env.reset()
while True:
    action = env.action_space.sample()
    next_obs, r, done, info = env.step(action)
    if done:
        print(info, env.current_step)
        break

Reference

• harfang3d/dogfight-sandbox-hg1