Raw Torque Data from Collision Experiments on a KUKA Robot (Part I: Accidental Collisions)

This dataset provides raw joint torque measurements from a KUKA LWR4+ robotic manipulator, recorded under accidental collision scenarios. It is primarily designed to support research in robot collision detection, classification, diagnosis, and prediction. How the Data Collected The data were collected in 2017 at the Chair of Automatic Control Engineering, Technical University of Munich, by Dr. Zengjie Zhang under the supervision of Dr. Dirk Wollherr. A detailed description of the experimental procedure and data acquisition protocol is provided in the following publication: [1] Zhang Z., Qian K., Schuller B. W., and Wollherr D. "An Online Robot Collision Detection and Identification Scheme by Supervised Learning and Bayesian Decision Theory." IEEE Transactions on Automation Science and Engineering, 2020, 18(3): 1144–1156. Dataset Overview The dataset comprises 206 collision sequences, consisting of raw torque sensor signals. All signals were sampled at **1 kHz** and are expressed in Newton-meters (Nm). Measurements are available for all seven joints (#1 to #7) of the KUKA robot arm. The data are stored in MATLAB (.mat) format. The files have been archived and compressed in batches of five for efficient storage and distribution. For instructions on extracting collision/contact segments or computing features from the raw signals, please refer to the accompanying GitHub repository. Quick-Start Option For users seeking a ready-to-use classification benchmark, a processed and pre-segmented version of this dataset is available as the quick-start dataset. The quick-start version is derived from the raw data presented here, but offers reduced flexibility in exchange for ease of use. Intended Use Cases This raw dataset offers maximum flexibility for advanced analysis and custom processing pipelines. Potential applications include, but are not limited to: - Development and benchmarking of collision detection algorithms - Supervised and unsupervised classification of contact events - Training and evaluation of predictive models for fault diagnosis - Design of active contact-reaction control strategies License and Citation MIT License. If you use this dataset or the associated work in your research, please cite both this dataset and the original publication [1] above.