AAU_Mehsed_UWB_Dataset

We recorded data from a UAV and stationary anchors in an indoor environment. UAV war equipped with a Pixhawk 4 flight controller and a companion computer (RPi4), an IMU, a barometer, and two UWB transceiver (Qorvo MDEK1001). The UAV flew within a volume of roughly 5x8x8 m^3, covered by a motion capturing system, tracking the pose of the UAV at 60 Hz with sub-millimeter and sub-degree accuracy for ground truthing.Nine stationary UWB nodes, each consisting of a UWB transceiver and a companion computer, were arbitrarily placed in the environment and their true positions was captured. A SDS-TWR protocol with a fully meshed ranging scheme/cycling was used to measure the distance between any device. A full raning cycle takes roughly 1 second. The net ranging rate is 100 Hz. The range measurements were streamed using the ROS1 middleware to the UAV to be recorded onboard in ROS1 bag files. The devices were synchronized with chrony.During 12 flights, the UAV took off to 2m, executed three upward spirals reaching 8m, and landed at its initial position. Anchors were randomly placed along the spiral flight path at heights ranging from 0.3m to 3.3m in an area of 5x8m^2. Each flight lasted around 140 seconds.The spatial relationships, the sensor noise characteristics, and the pair-wise biases between the intial set of anchors and tags were determined with our Python range evaluation package (https://github.com/aau-cns/cnspy_ranging_evaluation).Since the accuracy of TOA measurements is typically influenced by factors such as the spatial relationship between device antennas, pair-wise biases, and range-based biases, we substitute the actual range measurements of the 12 recorded flight datasets by the true ranges with Gaussian noise.For these synthetic (hybrid) datasets, the zero-mean white Gaussian noise was applied with different standard deviations (0.01, 0.1, 0.2, 0.3) and without biases, and with a different percentage of outliers (1% and 10%) and a measurement noise of noise sigma = 0.1. Note, that additional sequences can be created anytime, including biases, using the provided Python toolbox.

Further details can be found in the README.md