The Open-Source Dataset was organized in three subdirectories for Speed Estimation (Speed), Slope Estimation (Slope), and Mode Classification (Mode): Raw, Features, and Models. Raw: In the Raw directory, subject-specific comma-separated values (CSV) files contain multi-trial information that includes sensor data, mode and phase transitions, ground truth context values, and trial number. For Speed, context values have header values independent of other signals since speed was modulated by a Bertec Instrumented Treadmill. For Slope and Mode, context values are tied to mode transition headers since ramp ambulation occurred on static ramps. Foot IMU (Ay, Gx, and Gz), Shank IMU (Az, Gx, and Gy), Thigh IMU (Ay, Gx, and Gz), and Loadcell (Fy, Mx, and Mz) signals belonging to left-instrumented subjects (TF03, TF05, TF06, TF08, and TF09) were flipped to match signals of right-instrumented subjects. Features: The Features directory contains subject-specific CSV files in which each row corresponds to a row of features (mean, standard deviation, minimum, maximum, and last value) extracted from a set window of data (500 ms for Speed and Slope and 250 ms for Mode) at a consistent rate (50 Hz for Speed and Slope and 150 ms after every Early Stance for Mode). Mode, phase, and context labels are assigned to each row of features according to the mode, phase, and context value of the last value in the window from which the features were extracted. Models: The Models directory contains mode- and phase-specific XGBoost Speed and Slope Estimators and Mode Classifiers trained (90% training and 10% validation split) with all available data (denoted by “All” in the file name). XGBoost Error Estimators are included for Kalman filter implementations. An additional set of models are provided for the Baseline sensor suite outlined in Figure 6 (denoted by “Base” in the file name). These models can be embedded within the Raspberry Pi to predict/estimate in real-time.