Datasets

STINK Dataset

Data: 08.04.2019, Author: Hendrik Kolvenbach, DownloadDownload

The Sewer Terrain Inspection Dataset (STINK) was acquired with the legged robot ANYmal in the sewers of Zurich to assess the state of concrete deterioration. Sensors are placed in the feet of the robot to inspect the floor by perfoming a “scratching” motion. 18 sensor signals are recorded and evaluated (2x IMU and Force/Torque sensor). The condition of the concrete ranged from good to fair, while critical or extremely bad structural failures were not encountered. In total, we were able to collect 355 samples (good: 119 samples, satisfactory: 79 samples, fair: 157 samples) in different parts of the sewers, which were classified together with a professional sewer inspector who provided the ground truth. The dataset includes a Matlab-script to train a Support Vector Machine for classification. The dataset is linked to the following publication (external pagehttps://doi.org/10.3929/ethz-b-000351221).

Palpate Dataset

Data: 04.12.2018, Author: Hendrik Kolvenbach, DownloadDownload

The Planetary Soil Impact Dataset (PALPATE) is a dataset recorded to demonstrate the applicability of machine learning methods to classify the complex feet-soil interaction on fine grained soils. The dataset consists of 2600 feet-soil impacts, which were automatically executed and recorded with a specially designed testbed. Impacts were performed on a variety of Martian soil simulants, such as ES-1, ES-2, ES-3 and other. The dataset includes the sensor signals acquired by two different feet and sensors on the testbed. Additionally, a small dataset with 240 impacts created by the quadruped robot ANYmal is included. The dataset is linked to the following publication (external pagehttps://doi.org/10.1109/LRA.2019.2896732external page).