[Paper Review] INSANE: Cross-Domain UAV Data Sets with Increased Number of Sensors for developing Advanced and Novel Estimators

Christian Brommer, Alessandro Fornasier+5 more|arXiv (Cornell University)|Oct 17, 2022

Robotics and Sensor-Based Localization4 citations

TL;DR

This paper introduces INSANE, a cross-domain UAV data set with 18 sensors including multiple IMUs, GNSS, UWB, and cameras, designed for advanced localization and sensor fusion research. It provides centimeter-level RTK-GNSS and motion-capture ground truth across indoor, outdoor, and Mars-analog environments, enabling robust testing of long-term autonomy and novel estimation algorithms under real-world conditions.

ABSTRACT

For real-world applications, autonomous mobile robotic platforms must be capable of navigating safely in a multitude of different and dynamic environments with accurate and robust localization being a key prerequisite. To support further research in this domain, we present the INSANE data sets - a collection of versatile Micro Aerial Vehicle (MAV) data sets for cross-environment localization. The data sets provide various scenarios with multiple stages of difficulty for localization methods. These scenarios range from trajectories in the controlled environment of an indoor motion capture facility, to experiments where the vehicle performs an outdoor maneuver and transitions into a building, requiring changes of sensor modalities, up to purely outdoor flight maneuvers in a challenging Mars analog environment to simulate scenarios which current and future Mars helicopters would need to perform. The presented work aims to provide data that reflects real-world scenarios and sensor effects. The extensive sensor suite includes various sensor categories, including multiple Inertial Measurement Units (IMUs) and cameras. Sensor data is made available as raw measurements and each data set provides highly accurate ground truth, including the outdoor experiments where a dual Real-Time Kinematic (RTK) Global Navigation Satellite System (GNSS) setup provides sub-degree and centimeter accuracy (1-sigma). The sensor suite also includes a dedicated high-rate IMU to capture all the vibration dynamics of the vehicle during flight to support research on novel machine learning-based sensor signal enhancement methods for improved localization. The data sets and post-processing tools are available at: https://sst.aau.at/cns/datasets

Motivation & Objective

• Address the lack of open, high-fidelity, multi-environment UAV data sets for real-world localization research.
• Provide accurate, raw ground truth across diverse environments—indoor, outdoor, and Mars-analog—without filter-induced artifacts.
• Support research in cross-domain autonomy, including transitions between GNSS-available and GNSS-denied environments.
• Enable development of novel machine learning-based sensor signal enhancement and state-estimation methods through high-rate, vibration-sensitive sensor data.
• Facilitate benchmarking of visual-inertial, UWB, and multi-sensor fusion algorithms under realistic environmental effects such as EMI and signal degradation.

Proposed method

• Deploy a custom Micro Aerial Vehicle (MAV) equipped with 18 sensors, including dual RTK-GNSS, multiple IMUs (including high-rate MEMS), stereo cameras, UWB, and LRF.
• Use a dual-antenna RTK-GNSS setup with custom EMI shielding to achieve 1 cm (1-sigma) horizontal and 2 cm vertical position accuracy.
• Integrate a motion capture system for indoor ground truth and a fiducial marker field to bridge the outdoor-to-indoor transition gap.
• Record raw sensor data at high bandwidth using dual embedded platforms with four storage units (SD and SSD on each) to maximize data throughput.
• Apply marker-object association to align fiducial marker poses with RTK-GNSS and motion capture data, ensuring global and local ground truth continuity.
• Design the data acquisition system to balance computational load and fully utilize USB3 and sensor interface bandwidths.

Experimental results

Research questions

• RQ1 How can a unified, multi-environment UAV data set be created that supports cross-domain localization across indoor, outdoor, and planetary analog conditions?
• RQ2 What level of ground truth accuracy can be achieved using a combination of RTK-GNSS, motion capture, and fiducial markers in transition zones?
• RQ3 How do real-world sensor effects—such as EMI, signal dropouts, and vibration dynamics—affect state estimation and sensor fusion performance?
• RQ4 To what extent can high-rate IMU data improve the performance of machine learning-based signal enhancement and localization algorithms?
• RQ5 Can a standardized, open, and extensible data set format support long-term evolution and benchmarking of autonomous flight algorithms?

Key findings

• The INSANE data set provides sub-centimeter position accuracy (1 cm horizontal, 2 cm vertical) using a dual-antenna RTK-GNSS system with effective EMI shielding.
• Ground truth is available across three distinct environments: indoor motion capture, outdoor flight with GNSS, and Mars-analog outdoor flight, with seamless alignment via fiducial markers.
• High-rate IMU data (captured at 1 kHz) enables detailed analysis of vibration dynamics, supporting future research in sensor signal enhancement and noise modeling.
• The data set includes raw, unfiltered sensor measurements and ground truth, eliminating filter-induced artifacts and enabling more reliable algorithm validation.
• The system successfully recorded high-bandwidth data streams across four storage locations (SD and SSD on two embedded boards), achieving full utilization of available interface bandwidth.
• The data set is open-sourced and extensible, with plans to expand over time with new scenarios and sensor configurations, ensuring long-term utility for the research community.