A ten state quadcopter model.
To compute the MPC ingredients, run the offlinempcingredients.m file in MATLAB.
The output of the MATLAB file are already available in the folder mpc_parameters.
matlab -nodisplay -r "run('offlinempcingredients.m')"
The mpc parameters should be saved in human readable .txt form in the folder mpc_parameters.
To generate samples of the MPC, call
python3 samplempc.py sample_mpc \\
--numberofsamples=10
The results of this would be saved in a folder called datasets/quadcopter_N_{numberofsamples}_{date}-{time}.
You can similarly create a larger dataset, by calling this function in parallel
python3 samplempc.py parallel_sample_mpc \\
--instances=24 \\
--samplesperinstance=10 \\
--prefix=Cluster_test
The results of this would be saved in a folder called datasets/quadcopter_N_{instances*samplesperinstance}_merged_{prefix}_{date}-{time}
If you downloaded the precomputed dataset for this example, you should find it under datasets/quadcopter_N_9600000.
If you want to train an approximator on the precomputed dataset for this example, call
python3 approximatempc.py find_approximate_mpc \\
--dataset=quadcopter_N_9600000_merged_20221223-161206
The models will be saved in a models folder.
If you downloaded the pretrained NN, you should find it under models/10-200-400-600-600-400-200-30_mu=0.12_20230104-232806
You can run closed loop test with the model calling
python3 safeonlineevaluation.py closed_loop_test_on_dataset \\
--dataset=quadcopter_N_120000_test \\
--model_name=10-200-400-600-600-400-200-30_mu=0.12_20230104-232806 \\
--N_samples=3000 \\
--N_sim=1000