QCAT legged robot terrain classification dataset

Full description

The dataset includes measurements from two different sensor types on the quadruped robot DyRET. A 3-axis force sensor is mounted on the end of each of the four legs (Optoforce OMD-20-SH-80N), referred to as 'raw' in the dataset. The robot also has an Inertial Measurement Unit (IMU) mounted (Xsens MTI-30). It contains a 3-axis gyroscope providing rotational velocities, a 3-axis accelerometer providing linear accelerations, and a 3-axis magnetometer providing absolute orientation in reference to the Earth's magnetic field. The data is labeled 'imu' in the dataset. The robot walks forward on 6 different surfaces, available in the 'surface' column (0: Concrete, 1: Grass, 2: Gravel, 3: Mulch, 4: Dirt, 5: Sand). It does so at 6 different speeds, available in the 'speed' column (0-1: frequency 0.125 Hz; 1-2: frequency 0.1875 Hz; 3-4: frequency 0.25 Hz; 0,2,4: step length 80 mm; 1,3,5: step length 120 mm). There are 10 trials in each file (available in the 'eval_id' column) with 8 steps for each trial. This gives a total of 6*10*6*8 = 2880 steps in total. The jupyter notebook source code used for processing the force sensor data and IMU data is provided.

Lineage: The QCAT dataset was collected at different locations on CSIRO’s QCAT site in Brisbane, Australia, in November 2019 using the quadruped robot DyRET. The different environments comprising the data set are 1. Concrete road, 2.Grass, 3. Gravel, 4. Mulch, 5. Dirt path, and 6. Sand.
Data collection was done by walking with a fixed gait, but with three different step frequencies (0.125 Hz, 0.1875 Hz and 0.25 Hz) and two different step lengths (80 mm and 120 mm), for a total of six different speeds tested per surface. The robot walks forwards for ten trials of eight steps per speed and surface. A total of 2880 steps are available in the dataset. To be representative of the terrain type and reduce impact of local variation, each repetition was performed on a different part of the terrain.