We implement our algorithm and baseline algorithms on two models: inverted pendulum and quadrotor.

Environment:
    Python: 3.9.13
    Pytorch: 1.13.0
    NumPy: 1.26.0
    matplotlib: 3.8.0
    bayesian-optimization: 1.4.3
    tqdm: 4.65.0
    rowan: 1.3.0.post1

cd pendulum for inverted pendulum experiment.

To simply run the experiment under a certain setting, just run run.py for the complete training and testing task. You can also run the experiment under different settings. Examples:

# simply run
python run.py
# assign a wind condition
python run.py --wind breeze
# do not keep the log:
python run.py --logs 0

After the logs are created, you can run plot_result.py to plot the results in ./pics.

cd quadrotor for quadrotor experiment.

Similarly, run run.py for simple start. More settings are shown in following examples.

# simply run
python run.py
# do not keep the log
python run.py --logs 0
# assign a trajectory
python run.py --trace fig8
# assign a wind condition
python run.py --wind strong_breeze

Trajectory can be chosen from {fig8, hover, sin, spiral}. After the logentry is created, you can run plot_result.py --trace {fig8, hover, sin, spiral} to create 3D traces in ./traces and 2D projection in ./projections.

For both experiments, the argument parse --wind {breeze, strong_breeze, gale} is used to determine the wind velocity in the experiment.