RTISSLAM-3D: Full 6-DoF SLAM With an In-Air 3D Imaging Sonar

Jan Steckel; Vincent Katter; Wouter Jansen

cs.ROOct 15, 2026

RTISSLAM-3D: Full 6-DoF SLAM With an In-Air 3D Imaging Sonar

Authors: Jan Steckel, Vincent Katter, Wouter Jansen

Organizations: Cosys-Lab, Faculty of Applied Engineering, University of Antwerp, 2020 Antwerp, Belgium · Flanders Make Strategic Research Centre, 3920 Lommel, Belgium

Abstract

Echolocating bats navigate dense forests using sonar alone, and in-air 3D imaging sonars now capture the entire frontal hemisphere with every pulse. Sonar SLAM, however, has so far been planar: current systems estimate a 2D trajectory and discard the elevation information in the sonar images, an assumption that does not hold on rough and sloped terrain such as a forest floor. In this paper, we introduce RTISSLAM-3D, a novel 6-DoF SLAM system for a wheeled robot that carries an in-air 3D imaging sonar, an IMU and wheel encoders. A CFAR detector on the sphere reduces every sonar image to a sparse set of 3D detections, which we use for a 3D acoustic odometry that models the shape of the sonar beam, for place recognition with polar submap descriptors, and for the verification of loop closures by 3D registration, all combined in a 6-DoF pose graph with attitude factors. We evaluated RTISSLAM-3D on a 545m drive through a forest, with a 3D lidar SLAM as reference. The loop closures reduced the horizontal error of the acoustic odometry from 1.74m to 0.28m, with a height error of 0.12m, and 94.6% of the resulting 3D sonar map lies within 0.3 m of the lidar map.

Figures & tables

Panel of Figure 1 — Fig. 1: Overview of RTISSLAM-3D. Every eRTIS pulse is imaged over the frontal hemisphere (DMAS5 with coherence factor) and reduced to a sparse 3D detection set by a CFAR detector on the sphere; a fitted ground plane removes the ground multipath mirrors. The detection sets drive a 3D acoustic odometry (sonar translation, gyroscope rotation), keyframes every $1\text{\,}\mathrm{m}$ enter a 6-DoF pose graph with IMU attitude factors, and loop closures are proposed by a polar descriptor of the keyframe submaps and verified by 4-DoF submap registration. The 3D occupancy map is built on the optimized poses.

Submap	20 captures
Beam (range / az. / el.)	$3\text{\,}\mathrm{cm}$ / $3\text{\,}\mathrm{\SIUnitSymbolDegree}$ / $5\text{\,}\mathrm{\SIUnitSymbolDegree}$
Neighbors / gate	6 / $0.6\text{\,}\mathrm{m}$
Cov. floor / Cauchy scale	$5\text{\,}\mathrm{cm}$ / 2.5
Prior, rotation	$0.05\text{\,}\mathrm{\SIUnitSymbolDegree}$
Prior, fwd. / lat. / vert.	$1\text{\,}\mathrm{cm}$ / $5\text{\,}\mathrm{cm}$ / $5\text{\,}\mathrm{mm}$

TABLE I: Parameters of the 3D Acoustic Odometry (Prior per Step; Forward Prior Grows by

5\text{\,}\mathrm{\%}

of the Wheel Distance).

Fig. 2: 3D Detections of a Single eRTIS Capture in the Forest. Panel a) shows the view of the robot’s camera (for illustration only). Panel b) shows the acoustic image of the capture (DMAS5 with coherence factor, maximum over range, $40\text{\,}\mathrm{dB}$ dynamic range) over the frontal hemisphere as seen from behind the sensor, with circles at $30\text{\,}\mathrm{\SIUnitSymbolDegree}$ and $60\text{\,}\mathrm{\SIUnitSymbolDegree}$ off the sensor axis, and the directions of the CFAR detections (white circles; cyan crosses: ground mirrors). Panels c) and d) show the 3D detections in the sensor frame, in a side view and a top view, colored by their height above the fitted ground plane, on top of the lidar scan of the same instant (dark gray: lidar returns above the ground, light gray: ground). Red crosses: detections removed as ground multipath mirrors, i.e., more than $0.15\text{\,}\mathrm{m}$ below the fitted ground plane (dashed line in panel c)). The sonar sees the ground in front of the robot and the near sides of the trunks.

Fig. 3: 3D Acoustic Odometry Against the Lidar Reference. Panel a) shows the horizontal RPE (Relative Pose Error) as a function of the segment length, for the wheels and gyroscope alone (gray), the acoustic odometry in which the sonar estimates the full translation (green), and the final 3D acoustic odometry, in which the sonar estimates the horizontal translation only (orange). Panel b) shows the height RPE. Panel c) shows the dead-reckoning trajectories of the wheels and gyroscope and of the 3D acoustic odometry after a 4-DoF fit on the lidar reference (black). The sonar reduces the horizontal drift of the wheels by about

40\text{\,}\mathrm{\%}

, equally in both variants; estimating the vertical motion with the sonar reduces the height drift of the odometry alone, but not the error of the final SLAM trajectory (see text).

Fig. 4: 3D Place Recognition. Panel a) shows the polar descriptors (rings from 1 to $8\text{\,}\mathrm{m}$ , 60 sectors, the arrow marks the heading) of a query keyframe and of the keyframe it was matched with, for a loop closure with the median error. Panel b) shows the above-ground points of the two keyframe submaps after the 4-DoF registration (gray: matched keyframe, magenta: query). Panel c) shows the distribution (logarithmic scale) of the inlier fraction of the above-ground points after registration, for all top-five candidates of all keyframes without a position gate, split into candidates that were registered to within $0.5\text{\,}\mathrm{m}$ and $5\text{\,}\mathrm{\SIUnitSymbolDegree}$ of the reference (magenta) and the others (gray); the dashed line is the acceptance threshold.

Fig. 5: RTISSLAM-3D Against the 3D Lidar Reference. Panel a) shows the trajectories after a 4-DoF fit: the reference (black), the 3D acoustic odometry without loop closures (orange) and RTISSLAM-3D (blue); magenta circles mark the query keyframes of its 233 loop closures. Panel b) shows the height along the drive. Panels c) and d) show the horizontal and the height error of both trajectories on a logarithmic scale; the dashed lines are the RMSE of RTISSLAM-3D.

Variant	Loops	$xy$	$z$	Max	Bad
RTISSLAM-3D (ours)	233	0.28	0.12	0.64	0
No loop closures	0	1.74	2.82	3.23	–
Wheel odometry	237	0.81	0.09	1.46	0
Wheel odometry, no loops	0	2.40	3.00	3.87	–
Sonar vertical	217	0.32	0.23	0.69	0
Four height bands	236	0.28	0.12	0.66	0

TABLE II: Ablation of RTISSLAM-3D Against the Lidar Reference: Horizontal ATE, Height RMSE and Maximum Horizontal Error After a 4-DoF Fit (m), and Loop Closures With an Error Above

0.5\text{\,}\mathrm{m}

(Bad).

	Precision $\uparrow$		Recall $\uparrow$
Height Band	Lidar Poses	Ours	Ours
Ground ( $<$ $0.25\text{\,}\mathrm{m}$ )	0.970	0.959	0.363
Trunks, Vegetation (0.25– $2\text{\,}\mathrm{m}$ )	0.941	0.927	0.243
Canopy ( $>$ $2\text{\,}\mathrm{m}$ )	0.977	0.938	0.010
All	0.960	0.946	0.141

TABLE III: Precision and Recall of the 3D Sonar Map Against the Lidar Map, per Height Band (Tolerance

0.3\text{\,}\mathrm{m}

; Recall Within

6\text{\,}\mathrm{m}

in Front of the Sensor Poses).

Fig. 6: 3D Map of the Forest. Panel a) shows the 3D occupancy map of RTISSLAM-3D ( $0.1\text{\,}\mathrm{m}$ voxels) from above and panel b) the lidar map of the reference, limited to the points within $6\text{\,}\mathrm{m}$ in front of a sensor pose (the volume in which the recall is scored, Section IV ) and up to $3\text{\,}\mathrm{m}$ above the ground, at the same scale; the gray line is the path of the robot. Colors encode the height above the local ground; the ground itself is light gray. Panel c) shows the sonar map of the most revisited part of the site (dashed square) in perspective, and panel d) the same part from above, the sonar map above the ground drawn over the lidar map above the ground (gray). The trunks and the vegetation along the paths are resolved in the sonar map; the canopy is outside the sonar’s view.

Fig. 7: The Map and Trajectory as RTISSLAM-3D Builds Them (Three Moments of the Supplementary Video, Top View, Same Scale). The sonar map above the ground is carried by the keyframe poses of the estimate at that moment, the trajectory of RTISSLAM-3D is drawn in blue and the lidar reference in gray; magenta circles mark the loop closures found so far and the arrow the current pose. Panel a) shows the first lap, before any loop closure, panel b) the middle of the drive, and panel c) the final map.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Test	Loops	$xy$	$z$	Max	Bad
Inliers $\geq$ 0.75	290	0.27	0.13	0.62	1
Inliers $\geq$ 0.80	259	0.27	0.12	0.64	0
Inliers $\geq$ 0.85	233	0.28	0.12	0.64	0
Inliers $\geq$ 0.90	182	0.28	0.12	0.65	0
Inliers $\geq$ 0.95	95	0.28	0.11	0.67	0
Gate $3\text{\,}\mathrm{m}$	248	0.28	0.12	0.65	0

Appendix

TABLE IV: Sensitivity of RTISSLAM-3D to the Loop Closure Test (All Else as the Default, in Bold; Errors as in Table II ).

Simultaneous Localization and Mapping

Simultaneous localization and mapping estimates a robot’s motion while building a map of an unknown environment.

See all 193 papers

RTISSLAM: In-air Imaging Sonar SLAM using a Robust Pose Graph and Occupancy Grid Mapping
Oct 13, 2026Jan SteckelOccupancy Grid Mapping Simultaneous Localization and Mapping
Cosys-Lab, Faculty of Applied Engineering, University of Antwerp, 2020 Antwerp, Belgium · Flanders Make Strategic Research Centre, 3920 Lommel, Belgium
BatSLAM 2.0: Sequence-Verified Sonar Place Recognition in a Robust Pose Graph
Sep 30, 2026Jan SteckelSimultaneous Localization and Mapping Loop Closure Detection
Cosys-Lab, Faculty of Applied Engineering, University of Antwerp, 2020 Antwerp, Belgium · Flanders Make Strategic Research Centre, 3920 Lommel, Belgium
SOL-SLAM: Inverse Compositional Gauss-Newton Direct Registration for Fast Sonar-Only Local SLAM
Sep 16, 2026Kalvik Jakkala, Jason O'KaneSimultaneous Localization and Mapping Underwater Robotics
Texas A&M University

Oct 13, 2026J/K move · Enter open · S save

RTISSLAM: In-air Imaging Sonar SLAM using a Robust Pose Graph and Occupancy Grid Mapping

Jan Steckel

Cosys-Lab, Faculty of Applied Engineering, University of Antwerp, 2020 Antwerp, Belgium · Flanders Make Strategic Research Centre, 3920 Lommel, Belgium

Occupancy Grid Mapping Simultaneous Localization and Mapping Robot Localization

Open paper AskPDF arXiv

Sonar was one of the first sensors with which mobile robots built maps, and in-air sonar keeps working where cameras and lidar fail, in darkness, smoke, dust and next to glass. Classical sonar SLAM, however, relied on geometric features or occupancy grids built from sparse range readings, and both struggle with the specular reflections that dominate indoor sonar. In this letter, we introduce RTISSLAM, a novel sonar-only SLAM system for an embedded 3D imaging sonar, which combines the robust pose graph of BatSLAM 2.0 as back-end with an occupancy grid as map. Energyscapes from delay-multiply-and-sum beamforming with a range-wise CFAR serve as place descriptors for sequence-verified loop closures, acoustic odometry calibrates the wheel odometry online, recognized revisits are registered metrically, and the grid is stored per keyframe and anchored to the pose-graph nodes, so that it follows every loop closure. In simulated worlds with specular walls, RTISSLAM reduces the trajectory error of the wheel odometry from \SI{8.2}{\meter} to \SI{0.31}{\meter} in an office, while lidar-style grid SLAM on the same sonar data fails.

Abstract

Simultaneous Localization and Mapping

Explore similar work

RTISSLAM: In-air Imaging Sonar SLAM using a Robust Pose Graph and Occupancy Grid Mapping

BatSLAM 2.0: Sequence-Verified Sonar Place Recognition in a Robust Pose Graph

SOL-SLAM: Inverse Compositional Gauss-Newton Direct Registration for Fast Sonar-Only Local SLAM