PaperlayerView exact revision on arXiv

Source-linked AI summary

UWB-based system for UAV Localization in GNSS-Denied Environments: Characterization and Dataset

Jorge Peña Queralta, Carmen Martínez Almansa, Fabrizio Schiano, Dario Floreano, Tomi Westerlund

arXiv:2003.04380v2cs.ROcs.SYeess.SY

TL;DR

In GNSS-denied and indoor settings, UAV localization needs a practical alternative to costly, fixed motion-capture systems. The paper presents a portable UWB system and ToF-based dataset for UAV localization, then characterizes accuracy and anchor self-calibration. Its reported findings relate localization error to UAV speed, altitude, and anchor geometry, alongside initial autonomous-flight experiments.

  • Problem

    GNSS-denied and indoor UAV applications need localization, while motion-capture systems can be expensive and fixed and prior UAV datasets and analyses have limited anchor-setting coverage.

    Source: §I · p. 1, §II · p. 2

  • Method

    The paper studies a DWM1001 UWB system using ToF measurements, provides a UAV dataset and open-source calibration and ROS methods, and evaluates anchor self-calibration and localization across anchor and flight conditions.

    Source: §I · p. 2, B. UWB Localization Datasets · p. 3

  • Results

    Localization error was smaller at lower speed; with anchors at 1.8 m, higher flight altitudes gave smaller errors, while near-floor anchors showed no significant height differences.

    Source: D. Characterization of UWB… · p. 7

  • Takeaways & Limitations

    The dataset covers multiple anchor configurations and UAV tag counts, including autonomous flight with motion-capture ground truth, and characterizes accuracy by UAV speed, height, and distance from the anchor-system center of mass.

    Source: §V · p. 7

Abstract

from arXiv · show

Small unmanned aerial vehicles (UAV) have penetrated multiple domains over the past years. In GNSS-denied or indoor environments, aerial robots require a robust and stable localization system, often with external feedback, in order to fly safely. Motion capture systems are typically utilized indoors when accurate localization is needed. However, these systems are expensive and most require a fixed setup. Recently, visual-inertial odometry and similar methods have advanced to a point where autonomous UAVs can rely on them for localization. The main limitation in this case comes from the environment, as well as in long-term autonomy due to accumulating error if loop closure cannot be performed efficiently. For instance, the impact of low visibility due to dust or smoke in post-disaster scenarios might render the odometry methods inapplicable. In this paper, we study and characterize an ultra-wideband (UWB) system for navigation and localization of aerial robots indoors based on Decawave's DWM1001 UWB node. The system is portable, inexpensive and can be battery powered in its totality. We show the viability of this system for autonomous flight of UAVs, and provide open-source methods and data that enable its widespread application even with movable anchor systems. We characterize the accuracy based on the position of the UAV with respect to the anchors, its altitude and speed, and the distribution of the anchors in space. Finally, we analyze the accuracy of the self-calibration of the anchors' positions.

I. INTRODUCTION

The paper studies a portable, inexpensive UWB system for UAV localization in GNSS-denied and indoor settings, where motion-capture systems can be costly and fixed. It introduces ToF-based data and methods, characterizes localization and anchor self-calibration, and reports initial stable autonomous flight.

I. INTRODUCTION

  • The system is presented as a mobile, inexpensive alternative that can complement motion-capture systems with more flexible deployments and a relatively small positioning-accuracy impact.The paper also describes UWB as relevant to indoor robot navigation and aerial-robot tasks.Source: §I · p. 1
  • UWB positioning uses distance measurements between mobile tags and anchors at known positions, with distance estimated through ToF or TDoA.ToF estimates each anchor distance separately; TDoA requires connected anchors or accurate clock synchronization.Source: §I · p. 1
  • Prior UAV datasets and analyses relied on TDoA, whereas this work uses ToF and examines self-calibration and accuracy across UAV speed, height, and position relative to anchors.The paper states that this operational characterization was absent from previous works.Source: §I · p. 2
  • The paper contributes a ToF-based UAV localization dataset, anchor self-calibration analysis, accuracy characterization across anchor layouts and UAV conditions, initial autonomous-flight experiments, and open-source ROS and calibration code.The characterization varies anchor distribution, UAV distance from the anchors’ center of mass, speed, and height.Source: §I · p. 2

II. RELATED WORKS

Prior UWB localization studies cover mobile robots and UAVs, but existing UAV analyses and datasets leave limited evidence on fast ad-hoc deployments with movable anchors. The paper situates its work among mobile-robot uses and prior aerial-robot experiments.

II. RELATED WORKS

  • Previous UWB localization analyses used fixed, well-calibrated anchors, while existing datasets were small and specific, motivating study of fast ad-hoc deployments with movable anchors.The relative positions of such anchors may change over time.Source: §II · p. 2

A. UWB in Mobile and Aerial Robots

  • UWB has supported robot navigation, range-only SLAM, multi-robot relative localization, formation control, and coordinated UAV warehouse or mapping tasks.Some range-only SLAM settings can leave anchor positions unknown; multi-robot systems also use UWB with odometry or node-to-node distances.Source: A. UWB in Mobile and Aerial Robots · p. 2
  • A prior UAV docking setup used four anchors on a ground robot measuring 2 m by 1.5 m, while this dataset includes a four-anchor configuration for a roughly 0.6 m by 0.6 m robot.The dataset also includes a configuration with all four anchors near the ground.Source: A. UWB in Mobile and Aerial Robots · pp. 2–3

B. UWB Localization Datasets

Prior indoor-localization datasets cover limited settings or ranging tasks, while prior UAV data include a single flight and anchor setting. This work emphasizes varied UAV and anchor configurations and uses ToF measurements for positioning.

B. UWB Localization Datasets

  • Other datasets cover one anchor position with a tag at constant height, or inter-device and single-anchor ranging under LOS and NLOS conditions.These scopes differ from the UAV positioning configurations discussed in this paper.Source: B. UWB Localization Datasets · p. 3
  • Unlike a prior UAV dataset with one flight and one anchor setting, this dataset varies anchor configurations and UAV tag counts, includes planar anchors, and reports smaller out-of-the-box errors without EKF fusion.The authors note that planar deployments can mimic quick setups with anchors mounted on ground robots; one, two, or four tags also allow orientation estimation.Source: B. UWB Localization Datasets · p. 3
  • ToF-only positioning avoids the connected and synchronized anchors required by TDoA, which the authors identify as an enabler of ad-hoc mobile deployments despite potentially reduced accuracy.Table I is identified as a comparison with existing datasets.Source: B. UWB Localization Datasets · p. 3

III. METHODOLOGY

The methodology uses UWB two-way ranging to estimate signal time of flight (ToF) between unsynchronized devices. It describes both single-sided and double-sided ranging, with the latter adding an exchange for a more accurate ToF calculation.

III. METHODOLOGY

A. UWB Ranging with ToF measurements

B. Data Acquisition

The dataset was recorded through onboard-UAV and ground-station setups. It includes comparisons of existing datasets and anchor autopositioning methods, while the available passages do not provide the table values.

B. Data Acquisition

  • Dataset recordings used either an onboard computer on an autonomously flying quadrotor or a ground station recording a manually flown UAV.With a passive tag at the ground station, a delay occurs between motion-capture ground truth and UWB data.Source: B. Data Acquisition · pp. 3–4
  • The public dataset comparison is framed around existing UWB-based localization and positioning datasets.Source: Table I · p. 4
  • Table II compares Decawave DRTLS anchor autopositioning with a custom self-calibration method on latency and accuracy.Source: Table II · p. 4
  • The autonomous flight setup used an F450 quadrotor with a Pixhawk running PX4, an Intel Up Board running ROS, and a TF Mini lidar for height.Source: B. Data Acquisition · p. 4

C. Data Subsets

The dataset comprises seven subsets spanning anchor arrangements, tag counts, platforms, and flight regions. The paper also provides a custom anchor recalibration implementation to address slow and inaccurate RTLS autopositioning.

C. Data Subsets

  • Four anchors are considered the most suitable configuration for autonomous UAV flight, while six-anchor data examines vertical accuracy and intermittent anchor connectivity is also studied.Source: C. Data Subsets · p. 4
  • The seven subsets vary anchor placement and height, including a single-corner configuration and low, separated anchors tested for flights inside and outside the anchors’ convex envelope.The data also varies the number and arrangement of tags, including two tags 30 cm apart and four tags in a 20 cm by 30 cm rectangle.Source: C. Data Subsets · p. 4

D. Dynamic anchor reconfiguration

  • The custom anchor recalibration implementation uses 5 to 50 distance measurements and takes just under 1 s to 2.5 s, addressing RTLS calibration that takes around 40 s and can err by over 1 m.Source: D. Dynamic anchor reconfiguration · p. 4

E. Energy Efficiency

The paper monitors UWB-node power consumption across operating modes, noting its potential relevance when ground robots move only occasionally. The supplied passages also describe flight localization and error characterization across anchor geometry and ranging conditions.

E. Energy Efficiency

  • The UWB-node power consumption was monitored in different modes, with ground-robot energy use noted as relevant when robots move only from time to time.Source: E. Energy Efficiency · p. 4
  • In the autonomous flight demonstration, UWB estimates horizontal position, a 1D lidar estimates height, and Optitrack provides ground truth.Source: Figure 3 · p. 5
  • Position estimation is generally more accurate closer to the anchor-system center of mass, and errors are smaller at lower UAV speeds.Source: Figure 4 · p. 5
  • With four nearby anchors defining three faces of a cube, localization is highly unstable despite a small average error and relatively accurate estimates for half the measurements.Further filtering is needed for accurate flight in this configuration.Source: Figure 5 · p. 6
  • For calibrated line-of-sight ranging between two DWM1001 nodes, mean distance error is below 1 mm, standard deviation is 3.9 cm, and maximum error is 8.6 cm.Source: Figure 6 · p. 6

IV. EXPERIMENTATION AND RESULTS

The experiments test autonomous circular flight using UWB for horizontal localization and a 1D lidar for height. The waypoint control specifies the entry position, path tolerance, and angular speed.

IV. EXPERIMENTATION AND RESULTS

A. Flight with UWB-based localization

  • The waypoint controller uses a 0.3 m path threshold and 0.05 rad angular-speed increment, with entry position (1.23, 0, 1.23) in cylindrical coordinates.The waypoint is the objective position; the UAV's current position and the entry position are also represented in cylindrical coordinates.Source: A. Flight with UWB-based localization · p. 6
  • In the self-reconfigurable anchor, communication continues during calculation, making its power consumption greater than Decawave’s DRTL setup.The tags were powered through USB in all measurements.Source: Figure 7 · p. 6

B. Spatial distribution of anchors

Anchor placement affects localization accuracy: the study compares a conventional room-corner arrangement with compact and movable-platform configurations. In the room-corner flight, most samples had errors below 10 cm.

B. Spatial distribution of anchors

  • Over 50% of room-corner flight samples had error below 10 cm, and the x-error lower and upper quartiles were below 5 cm.The paper notes this flight reflects a typical arrangement with anchors at room corners.Source: B. Spatial distribution of anchors · p. 6
  • With four anchors 60 cm apart at cube corners, position estimates were highly unstable despite relatively low lower and upper quartile errors.The authors suggest filtering and sensor fusion might make this arrangement usable when the error margin allows.Source: B. Spatial distribution of anchors · p. 6
  • For four near-ground anchors separated by around 2 m, localization accuracy doubled compared with the preceding configuration.The authors describe this distribution as the most usable one in the dataset for a moving platform.Source: B. Spatial distribution of anchors · p. 6

C. Autopositioning of anchors

The paper presents a self-calibration procedure for anchor positions and characterizes localization accuracy across anchor layouts, UAV distance, speed, and height. The measurements show that accuracy depends on these operating conditions.

C. Autopositioning of anchors

  • The authors report better results from Decawave’s UART API localization than from raw distance measurements processed by open-source multilateration algorithms.They also report Decawave’s anchor autopositioning took around 40 s and did not give good results in their experience; their firmware recalculates anchor positions after movement.Source: C. Autopositioning of anchors · p. 7
  • More than 50% of self-calibration cases had overall error standard deviation under 3 cm across measurements at over 35 distances up to 22 m.At individual distances, the standard deviation ranged from 0 to 4 cm; each distance had 50 measurements.Source: C. Autopositioning of anchors · p. 7
  • The initial autopositioning firmware consumes very high power because nodes transmit at high frequency during the procedure.The authors describe this implementation as not power-optimized; each anchor takes an initiator turn and calculates distances to the others using two-way ranging.Source: C. Autopositioning of anchors · p. 7
  • The self-calibration procedure assumes the counter-clockwise ordering of anchors around the convex-envelope boundary and defines the x-axis from the first to second anchor.The first anchor activated as initiator is treated as the coordinate origin.Source: C. Autopositioning of anchors · p. 7

D. Characterization of UWB localization accuracy

  • Position estimates are more accurate when the UWB tag is closer to the anchor system’s center of mass.This pattern is reported for both room-corner anchors and anchors near the floor at the center.Source: D. Characterization of UWB… · p. 7
  • Lower UAV speed corresponds to smaller position-estimation error; height effects depend on anchor placement, with no significant difference for near-floor anchors but smaller errors at higher flight altitudes for room-corner anchors at 1.8 m.With line of sight maintained, error was smallest near the constant z-plane defined by anchor positions.Source: D. Characterization of UWB… · p. 7
  • The accuracy findings can inform deployment control strategies that adjust error estimates based on speed, height, and distance to the anchor-system center of mass.The paper frames this as a way to support more efficient autonomous-UAV control in real deployments.Source: D. Characterization of UWB… · p. 7

V. CONCLUSION AND FUTURE WORK

The paper presents a UWB localization dataset for aerial robots, covering multiple anchor configurations, variable tag counts, and autonomous flight, with Optitrack ground truth. It also identifies future work on multi-UAV and distributed localization and UWB fusion with odometry.

V. CONCLUSION AND FUTURE WORK

  • The dataset includes multiple anchor configurations, variable numbers of UAV-mounted UWB tags, and autonomous-flight data, with Optitrack ground truth in all cases.The authors describe it as the largest and most complete dataset of its kind to their knowledge.Source: §V · p. 7
  • The conclusion identifies the dataset as supporting further exploration of UWB localization for autonomous flight in GNSS-denied settings with UAVs and reference ground robots.The excerpt describes this as the authors’ expectation for the dataset’s use.Source: §V · pp. 7–8
  • Future work will study UWB for robust multi-UAV systems, anchor-free distributed localization using mesh networks and relative positions, and fusion with inertial and visual odometry.Figure 8 shows data from four tags on one rigid body and four anchors centered in the motion-capture arena, separated by 1.8 m; Figure 9 plots error probability against distance to anchor-system center and height.Source: §V · p. 8, Figure 8 · p. 8, Figure 9 · p. 8
Loading 2003.04380v2…