dynamics_model_learning

A repo for CMU's 16-715 Advanced Robot Dynamics course project

Data Loader Instructions

1. Download the processed data from the dataset online and put it in the top level directory of this project (dynamics_model_learning/)
2. Import the DataLoader class (make sure your script is in top directory - dynamics_model_learning): from data.DataLoader import DataLoader, DynamicsDataset
3. Initialize it with the path to the processed data from the dataset: DL = DataLoader(path)
4. Load in some data by doing one of the following: - Use pre-selected set of data: DL.load_easy_data() - Hand-pick the flights you want from flights_info.txt:
   1. Create a list of the flights you want to load (ex: "2021-02-03-13-44-49"): selected_data = ["2021-02-03-13-44-49", "2021-02-03-13-44-49"]
   2. Tell the data loader to load in the selected data: DL.load_selected_data(selected_data)
5. To save the data to an npz file, do: DL.saveData(path), where path is the path to the file you want the data to be saved in (ex: model/train_data.npz) - The data can be loaded into a file using: npzfile = np.load(filePath) - The state data can be accessed with: npzfile["input"] - The control inputs (RPM values) can be accessed with: npzfile["control_inputs"] - The state derivative can be accessed with: ``
6. You can now train a network using this data, see dynamics_model_learning/model/train.py for an example on how to do that.
7. (optional) Load the data into a torch.utils.data.Dataset function by doing: torchDatasetObject = DynamicsDataset(DL.get_state_data(), DL.state_dot_values)
8. (optional) Get chunks of the data as with the following functions:
   • DL.get_column_names() will return a list of the coumn names (i.e. state and input variables)
   • DL.get_time_values() returns [N] size numpy array of the time values for each data point
   • DL.get_state_data() returns [N x 23] numpy array of the state data NOTE: includes rpm values as well (part of state but also input to system)
   • DL.get_control_inputs() returns [N x 4] numpy array of rpm values (in order 1 -> 4 that fits paper's dynamic model, check Fig 3 in the paper for exact numbering scheme)
   • DL.get_des_rpm_values() returns [N x 4] numpy array of desired rpm values at every time step (used to solve for rpm_dot = DL.motor_time_constant * (desired_rpms - current_rpms) )
     • from eq 20 on page 7 of paper
   • DL.get_battery_voltage_data() returns [N] size numpy array of battery voltages during flights