Machine learning for unicycle riding
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
gym-unicycle
trained
.gitignore
README.md
env.py
main.py
makeEnv.sh
requirements.txt
testAgent.py

README.md

Machine Learning For Unicycling

This project is for training a neural network to ride a sumlation of a unicycle, using reinforcement learning.

Video of simulated unicycle riding

I took the cartpole example from here, and modified to be a unicycle.

I find the gym+keras ecosystem terribly confusing. Having a custom environment is really confusing, in terms of directory structure. So the env.py is supposed to be a shortcut to gym-unicycle/gym_unicycle/envs/unicycle_env.py. But sometimes shortcuts don't work well with git.

How to Use

  • Run ./makeEnv.sh. This creates a virtual env. I can't have just a requirements.txt file, because the installation of gym and a custom env is a bit odd.