jinPrelude / kerbal-rl

Reinforcement learning environment for Kerbal Space Program

Kerbal-rl

kerbal-rl is aerospace reinforcement learning environment using Kerbal Space Program.

Environments

Currently available environments are :

• hover-v0 : Given the highest score if the rocket is hovering at the target altitude. target altitude is given randomly at avery episodes.

• hover-v1 : Same feature with hover-v0, but the sparse reward is given.

Installation

To use this package, you must have Kerbal Space Program with krpc mod installed.

• For more information on Kerbal Space Program please visit here.

• For more information on krpc please visit here.

This code will automatically install the package.

git clone https://github.com/jinPrelude/kerbal-rl.git
cd kerbal-rl
pip install .

Prerequisite

1. Run KSP, go to vehicle assembly building, and make your own rocket. If you are not familiar with making a rocket, use premade vehicle 'test_vehicle.craft' which is involved in our repo by following the instruction below :

   1. Copy test_vehicle.craft file.
   2. Move directory to ksp local folder.
   3. Paste the file in Ships\VAB folder.
   4. Run KSP. You could find the 'test_vehicle' in the game.

2. Click the launch button to go to the launch pad.

3. Lastly, Click the button 'Start Server' on kRPC window. If there's no any windows named kRPC, check our if the krpc mod has successfully installed.

Let's get back to the code. You can simply import & initialize kerbal-rl like :

import kerbal_rl
env = kerbal_rl.make('hover-v0')
env.reset()

You could feel more familiar with kerbal-rl if you have experience using Openai gym.

How it works

kerbal-rl Capture the scene for every 0.1 second for a step. It means, Although your computing time for desicion making takes time more than 0.1 second, ksp doesn't wait for it. Fortunately, There is the hyperparameter 'interval', that allows you to adjust so you could fine-tuning the most appropriate interval.

import kerbal_rl
env = kerbal_rl.make('hover-v0')

# You can customize your environment like :
env(max_step=100, max_altitude=10, sas=True, interval=0.1)
env.reset()