This code demonstrates how to train optimal state-feedback controllers for a 2D quadcopter model represented by a feedforward neural network.

The properties of these controllers are investigated in the following papers:

• Tailor, D., Izzo, D. (2019). Learning the optimal state-feedback via supervised imitation learning. Astrodynamics (Springer).
• Izzo D., Tailor D., Vasileiou T. (2019). On the stability analysis of optimal state feedbacks as represented by deep neural models. In Review

This code has only been tested on a UNIX environment with Python 3.6. We use virtualenv with pip for package management.

The code responsible for trajectory optimisation requires ampl and snopt executables to be in the system path. Datasets are provided so the rest of the code is usable if these are unavailable.

The environment variable needs to be set: PYTHONHASHSEED=0 as the code responsible for exporting and reading trained keras models depends on reproducible hashes. It is easiest to add this to the activate script of virtualenv where the variable can be unset also on deactivation.

Package PyDDE cannot be installed with pip and must be installed from source:

git clone https://github.com/hensing/PyDDE.git
cd PyDDE
python setup.py install