Optimal closed loop control through DNN

Optimal control problems (OCP) can only, in general, be solved in open loop, i.e. with the control and states as a function of time.

This project aims to learn a function that maps from the states to the control "optimally" trought Deep Neural Network (DNN).

The OCP is solved using PSOPT and the DDN using TensorFlow.

Docker

There is a dockerfile in the docker folder that creates an enviroment to run PSOPT.

Build

To build the container on your local machine named as optdnn_env, cd to /docker and use:

docker build . -t optdnn_env 

Note: don't forget the dot.

Run

To run something in the environment, for example, the example launch of PSOPT, you can use:

docker run -it --rm  optdnn_env:latest ./psopt/build/examples/launch/launch 

Note: To compile your own example you should implement it in the file /PSOPT/examples/user/user.cxx. Then, cd to /PSOPT/build/examples/user and run

make

This will create your compiled file in /PSOPT/build/examples/user.

Run inside the docker container

To run inside the docker container we will have to bind mount two folders with the container, namely, /build to ~/build/examples/user and /src to ~/examples/user, this means that you will have to place your .cxx file inside /src and then, to run, just run in your terminal the following command:

docker run -it --rm -v $PWD/src:/usr/src/optdnn/psopt/examples/user/ -v $PWD/build:/usr/src/optdnn/psopt/build/examples/user optdnn_env:latest bash

In this command -u$(id -u):$(id -g) maps your user to the docker container user, -it runs interactively, --rm deletes the container after you leave, -v $PWD/src:/usr/src/optdnn/psopt/examples/user/ bind mount /src to /examples/user, -v $PWD/build:/usr/src/optdnn/psopt/build/examples/user bind mount /build to /build/examples/user, optdnn_env:latest is the image and bash runs the bash.

With this your implemented .cxx file is already inside /psopt/examples/user/ and you can just cd to /psopt/build/examples/user and run make and then the user executable.

Run TensorFlow in Docker

To run tensorflow inside the docker container use the following command

docker run --gpus all -it --rm -v $PWD:/tmp -w /tmp optdnn_env:latest python ./script.py