/RLFluidControl

reinforcement learning for active fluid control

Primary LanguageCMIT LicenseMIT

RLFluidControl

Reinforcement learning for active fluid control.

See our paper: Reinforcement Learning for Active Flow Control in Experiments (https://arxiv.org/abs/2003.03419)

Dependencies

Python Packages: TensorFlow 1.x (1.13+ suggested), numpy

How to run

Start the server (RL agent) by

$ cd server
$ python server.py -env CFD -fil None

-fil None means that we don't use any filter, in consistent to our practice for the CFD environment in the paper.

Start the client:

2D BDIM:

We provide a simple example using Lilypad, a code based on a 2D BDIM method developed by Dr. Gab Weymouth. See https://github.com/weymouth/lily-pad.

Run clientLilypad/clientCFD.pde in Processing (download from https://processing.org/download/).

3D LES:

See details in clientNektar/Readme