OpenAI Gym environment for reinforcement learning with multicopters.
-
Pure Python / Cross-platform
-
Uses realistic multirotor dynamics (Bouabdallah et al. 2004) that can be subclassed for a particular vehicle configuration (quad, hex, octo, etc.)
-
Supports rendering via a Heads-Up Display (HUD) similar to Mission Planner / QGroundControl.
% pip3 install gym
% python3 setup.py install
% python3 gym_copter/envs/lander2d.py
(On Linux you will probably have to run pip3 with sudo
You should see the copter land safely.
The NEAT sub-folder of this repository shows how you can use the NEAT algorithm to evolve a neural controller for your copter.
-
Lander-v0 2D LunarLander-style challenge
-
Lander3D-v0 3D lander with reward proportional on proximity of touchdown to center
-
Lander3DHardcore-v0 3D lander with big reward for landing inside circle