GymArt

Additional environments for the OpenAI Gym.

Currently contains the following environments:

• QuadrotorEnv. This is an environment for quadrotor stabilization at the origin. The env has static episode length that depends on episode time in sec ep_time, sim_freq that is integration freq, and sim_steps - number of simulation time steps after action is applied. For example: for ep_time=5, sim_freq=200 and sim_steps=2 the environment will have 500 steps.

Requirements

• python 3.X

Installation

Clone it:

mkdir ~/prj
cd ~/prj
git clone https://github.com/amolchanov86/gym_art.git

Add it to your .bashrc

export PYTHONPATH=$PYTHONPATH:~/prj/gym_art

Experiments

Test default quadrotor with the Mellinger nonlinear controller

cd ~/prj/gym_art/gym_art/quadrotor
./quadrotor.py

Test CrazyFlie quadrotor with the Mellinger nonlinear controller

cd ~/prj/gym_art/gym_art/quadrotor
./quadrotor.py -q crazyflie

see --help option if you want to experiment with other parameters.

Remarks:

• I tested everything with the Garage's anaconda environment. Hence, please, install Garage if something goes wrong.
• The current default version of the quadrotor model is roughly approximates AscTech Hummingbird quadrotor. All supported models:
  • Hummingbird. Parameters were borrowed in large from the Rotors Simulator. See Hummingbird URDF for more details.
  • CrazyFlie