See src/setup.txt to set up the Python environment.
To reproduce results, any of the trained models can be used by running the notebook files. All results included in the paper can be obtained by running the respective notebook file. Interesting ones may be the ___ Wind Impact.ipynb files and Plot Trajectories.ipynb files.
To train your own models or reproduce our trained models, make the appropriate changes to scripts/opt_multirotorenv.py and run that file as denoted in src/README.txt. These models will take about a day to arrive at the same results as shown in the paper on a modern GPU with 10 parallel processes. To train the LSTM model, the training data will linked here shortly.
See Demo.ipynb for basic usage of the UAV without any disturbances.
The final hyperparameters for each of the models can be seen by running optuna dashboard sqlite:///filename.db from the studies/ directory where filename is the model you want to inspect the hyperparmeters. In this dashboard, you can see the impacts of each hyperparmeter along with other useful information. The final models chosen are the models with the highest value in this dashboard.