VFRN4UAVs

Using Docker

Building

docker build -t vfrn .

Running

docker run -it --gpus=all --net=host --env DISPLAY=$DISPLAY vfrn

Post-run

When using vscode dev-containers verify if the cpu is going to 100%, if so try this inside the container:

ps aux | grep @vs

and kill the process that appears in the list

Running the project

Using conda env "drones":

conda activate drones

Running the train of the UAV (at the moment, this is using mysql to run multiple iterations in optuna and there is the need to create the db before this can be run)

python /app/gym-pybullet-drones/gym_pybullet_drones/examples/learn.py --gui false --record_video false

Visualize the control

python /app/gym-pybullet-drones/gym_pybullet_drones/examples/just_visualize.py --gui false --record_video false

Generate the dataset

python /app/gym-pybullet-drones/gym_pybullet_drones/examples/dataset_generator.py --x 0 --y <starting_point_y> --z <starting_point_z> --step <step (negative value, if it starts in the right or up part)> --step_axis <axis that is used with step for dataset record>

Generate vnnlib and vis dataset

python /app/gym-pybullet-drones/gym_pybullet_drones/examples/dataset_generator.py --obs kin --act discrete_2d_complex --dataset_folder <dataset_folder, ex: /app/VFRN4UAVs/discrete/datasets/dataset_31.08.2024/19.14.17> --perturb <perturbation radius, ex: 0.01> --output_condition quadrants --visualize_dataset True --visualize_actions True

Activate alpha-beta-crown env

conda activate alpha-beta-crown

Launch alpha-beta-crown verification

python /app/alpha-beta-CROWN/complete_verifier/abcrown.py --config /app/VFRN4UAVs/discrete/abcrown_yaml/onnx_sac_discrete_multiple_vnnlibs_reward440.yaml

Activate the conda env drones again!

conda activate drones

Plot the certified and violated points in alpha-beta-crown

python /app/VFRN4UAVs/plot_certification.py --file_path /app/certified.txt --directory /app/VFRN4UAVs/discrete/datasets/dataset_31.08.2024/final --include_unsafe True

Plot the counter examples and the action

python /app/VFRN4UAVs/plot_cex.py --directory /app/VFRN4UAVs/discrete/datasets/dataset_31.08.2024/final/vnnlibs_0.01_quadrants

Activate the conda env pynever

conda activate pynever

Verify with pynever. the file exp_launcher needs to be changed for new datasets

python /app/VFRN4UAVs/pyNever/exp_launcher.py 1 60

Activate the conda env drones again!

conda activate drones

Plot certified in pynever

python /app/VFRN4UAVs/pyNever/exp_launcher.py --results_path /app/VFRN4UAVs/pyNever/logs/experiments_too_tilted_436.csv --directory /app/VFRN4UAVs/discrete/datasets/dataset_20.08.2024/final