Learning to walk with genetic algorithms by optimizing a truncated Fourier series, applied to a realistic bio-mechanical model of the human body.
It used the OpenSim environment for biomechanical simulations, modified for the NIPS 2017 "Learning to Run" competition. Here you can find their repository with a detailed introduction: https://github.com/stanfordnmbl/osim-rl.
Here you can find an article about this project where I explain it in some detail. https://medium.com/towards-data-science/learning-to-walk-with-evolutionary-algorithms-applied-to-a-bio-mechanical-model-1ccc094537ce