M-CROSS: A Multisensor Simulation Dataset for SLAM across Water--Land Transitions
Authors: Yan Dong, Tianyi Wu, Junyu Yang, Shaoqiang Qiu, Zefeng Yan, Tao Zhang, Bin Han
Organizations: School of Automation, Tsinghua University, Beijing, China · State Key Laboratory of Intelligent Manufacturing Equipment and Technology, School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan, China · School of Automation and Software Engineering, Shanxi University, Shanxi, China
Water--land cross-domain robots require continuous localization across land, transition and underwater phases, where sensor characteristics and availability change abruptly. Existing datasets and conventional metrics are not designed to capture transition-induced degradation and failures. We present M-CROSS, to the best of our knowledge, the first multisensor simulation dataset and benchmark for SLAM across continuous water--land transitions. It provides synchronized LiDAR, sonar, stereo camera, IMU, DVL, and depth measurements; 6DoF pose ground truth; sensor-wise medium-state annotations; and medium-dependent sensor degradation models. We establish a transition-aware protocol comprising the Trajectory Completion Rate (TCR), Piecewise Absolute Pose Error (PW-APE), Transition End Error (TEE), and Transition Survival Period (TSP), which measure trajectory completion, single-domain accuracy, transition-induced error, and localization survival under prolonged partial immersion, respectively. Experiments with 6 representative SLAM systems show that existing methods fail to maintain both localization accuracy and trajectory continuity across the full set of water--land transitions, exhibiting transition-induced drift, trajectory discontinuities, and tracking failures. M-CROSS is publicly available https://anonymous.4open.science/r/MCross-Dataset
Figures & tables
Fig. 1: Overview of the M-CROSS simulation scene. The cross-domain robot traverses from one riverbank to the other through continuous land–water transitions. Representative LiDAR, camera, and sonar measurements are shown at the bottom.
Fig. 2: Overview of the dataset construction and benchmarking workflow.
Sensor
Rate
Key Configuration
Stereo RGB
20 Hz
1280×720 , 0.107 m baseline
LiDAR
10 Hz
32 channels, 360∘×30∘ FoV
Sonar
5 Hz
130∘×20∘ FoV, 0.5–10 m range
Depth Sensor
5 Hz
0–16 bar, 0.01 m depth noise std.
DVL
5 Hz
0.05–125 m bottom-track range, 0.005 m/s velocity noise std.
IMU
200 Hz
60μg/Hz acc., 0.007∘/s/Hz gyro. noise density
TABLE I: Key sensor parameters of M-CROSS
Seq.
Domain Pattern
Traversal Description
Dur. (s)
Len. (m)
seq01
LTU
Nominal land-to-underwater entry over flat terrain, with rocks and a submerged wreck
39.32
45.59
seq02
UTL
Nominal underwater-to-land exit over flat terrain, with rocks and a submerged wreck
39.89
45.05
seq03
LTU
Challenging land-to-underwater entry over rough terrain, with rocks, vegetation, a floating boat, and swimming fish
61.22
69.28
seq04
UTL
Challenging underwater-to-land exit over rough terrain, with rocks, vegetation, a floating boat, and swimming fish
86.54
64.40
seq05
LTUTL
Land–underwater–land loop over rough terrain, with trees, rocks, vegetation, a floating boat, and swimming fish
103.88
117.69
seq06
UTLTU
Underwater–land–underwater loop over rough terrain, with trees, rocks, vegetation, a floating boat, and swimming fish
87.38
101.90
TABLE II: Overview of the M-CROSS dataset sequences and their cross-domain traversal patterns.
Fig. 3: Overview of the M-CROSS environment and sequence trajectories. The central panel shows the overall simulation environment and the trajectories of all eight sequences, while panels (a)–(e) present representative views of different regions and environmental elements. Star and triangle markers indicate the starting and ending points of the trajectories, respectively. For each sequence, trajectory segments above the water surface are shown as solid lines, whereas underwater segments are shown as dashed lines.
Fig. 4: Definition of cross-domain phases. A completed Transition spans the first body–surface intersection ( tts ) to full entry into the target medium ( tte ); an incomplete crossing remains in the Transition phase until the annotated endpoint tc .
Seq.
Fast-LIO2 [ 25 ]
Bruce-SLAM [ 26 ]
ORB-SLAM3 [ 27 ]
VINS-Fusion [ 28 ]
LVI-SAM [ 29 ]
Fast-LIVO2 [ 30 ]
seq01
36.75% (U 1 )
0.00% (I)
9.48% (L 1 )
100.00%
48.53% (U 1 )
28.18% (T 1 )
seq02
12.90% (U 1 )
77.39% (T 1 )
16.91% (U 1 )
17.36% (U 1 )
10.19% (U 1 )
0.00% (I)
seq03
0.00% (I)
0.00% (I)
20.53% (T 1 )
100.00%
57.17% (U 1 )
24.75% (T 1 )
seq04
4.58% (U 1 )
0.00% (I)
8.73% (U 1 )
100.00%
10.28% (U 1 )
0.00% (I)
seq05
13.84% (U 1 )
0.00% (I)
5.92% (T 1 )
100.00%
13.37% (U 1 )
9.19% (T 1 )
seq06
5.62% (U 1 )
29.35% (U 1 )
20.25% (U 1 )
100.00%
4.47% (U 1 )
0.00% (I)
TABLE III: Trajectory completion rate of six SLAM methods on all eight M-CROSS sequences.
Fig. 5: Trajectory comparison of the evaluated SLAM methods on all eight M-CROSS sequences. Each subfigure shows top-down 2-D projections of the estimated and ground-truth trajectories. Star and triangle markers indicate the starting and ending points of the trajectories, respectively.
Method
Single-domain performance
Transition-phase performance
Valid SD-Seg. ∗ ↑
PW-APEP/R (m/ ∘ ) ↓
Valid Trans. ∗ ↑
TEEP/R (m/ ∘ ) ↓
Fast-LIO2 [ 25 ]
7
0.57 / 1.03
1
1.03 / 2.07
Bruce-SLAM [ 26 ]
1
0.56 / 0.78
0
N/A / N/A
ORB-SLAM3 [ 27 ]
4
0.69 / 0.54
1
3.10 / 21.21
VINS-Fusion [ 28 ]
19
1.03 / 1.17
12
0.25 / 0.69
LVI-SAM [ 29 ]
9
0.96 / 1.10
2
1.80 / 0.30
TABLE IV: Dataset-level summary of PW-APE and mean TEE on M-CROSS.
Method
TSP 0.5
TSP 1
TSP 2
Fast-LIO2 [ 25 ]
7.59
33.90
33.90
ORB-SLAM3 [ 27 ]
0.10
0.33
1.20
VINS-Fusion [ 28 ]
0.87
1.41
1.49
LVI-SAM [ 29 ]
3.96
4.02
4.14
Fast-LIVO2 [ 30 ]
0.87
1.74
3.21
TABLE V: TEP under different position-error thresholds on seq08 .
Fig. 6: Comparison of vertical-position estimates during a land-to-underwater transition. (a) Methods relying on a single primary exteroceptive sensor exhibit substantial drift or tracking termination as the sensing medium changes. (b) Camera–LiDAR fusion methods maintain better trajectory continuity and reduce the localization error, but noticeable deviations remain during the transition and underwater phases. The shaded regions indicate the land, transition, and underwater phases.
Multi-robot simultaneous localization and mapping (SLAM) is a fundamental task in multi-robot operations. Robots must have a common understanding of their location and that of their team members to complete coordinated actions. However, multi-robot SLAM between Uncrewed Surface Vessels (USVs) and Autonomous Underwater Vehicles (AUVs) has primarily been achieved through acoustic pinging between robots to retrieve range measurements; a measurement technique requires that robots to be in similar locations simultaneously, have an uninterrupted path for signal propagation, and may necessitate synchronized clocks. This is especially challenging in complex, cluttered maritime environments, where structures may impede signals. However, these same structures may be observable above and below the water's surface, presenting an opportunity for inter-robot SLAM loop closure between USV and AUV data streams. This work builds upon recent research on inter-robot SLAM loop closure between USV and AUV data, extending it to propose a centralized multi-robot SLAM system. Each robot performs its state estimation, and we detect loop closures between each AUV and the USV data. These inter-robot loop closures are used to merge each robot's state estimate into a centralized graph, yielding estimates for the whole time history of the USV and all AUVs in the system. Validation is performed using real-world perceptual data in three different environments. Results show improved errors for AUVs in the multi-robot SLAM system compared to single-robot SLAM over the same trajectories. To our knowledge, this is the first instance of a multi-robot SLAM system with AUVs and USVs built on loop closures rather than acoustic distance measurements.
John McConnell, Armon Shariati, Paul Szenher +1
United States Naval Academy, Annapolis, MD, USA · Shield-AI, San Diego CA, USA · Stevens Institute of Technology, Hoboken, NJ, USA
Underwater robots typically use both cameras and sonar for perception to leverage the rich semantic details of vision and the robust range measurements of acoustics. However, learning to map between these modalities via cross-modal prediction remains underexplored due to limited sonar-visual paired datasets. We present SOVIS, a sonar-visual dataset for cross-modal underwater perception. SOVIS comprises over 76,000 paired frames collected across 17 dives at six sites in the Trondheimfjord, supported by an end-to-end pipeline that cleans and synchronizes the cross-modal sensor data. We also introduce an interactive annotation tool designed to accelerate the labeling process for this paired data. Finally, we demonstrate a proof-of-concept cross-modal fish detection task using a small subset of labeled data, achieving a 7x improvement in [email protected] over a monocular camera baseline. SOVIS serves as the first step toward advancing cross-modal underwater perception research, enabling research directions such as dense sonar prediction from monocular images.
Weitung Chen, Phil Tinn, Per Gunnar Auran +2
Massachusetts Institute of Technology (MIT), Cambridge, MA, USA · SINTEF, Trondheim, Norway · Department of Marine Technology, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
Recent years have seen a focus on research into distributed optimization algorithms for multi-robot Collaborative Simultaneous Localization and Mapping (C-SLAM). Research in this domain, however, is made difficult by a lack of standard benchmark datasets. Such datasets have been used to great effect in the field of single-robot SLAM, and researchers focused on multi-robot problems would benefit greatly from dedicated benchmark datasets. To address this gap, we design and release the Collaborative Open-Source Multi-robot Optimization Benchmark (COSMO-Bench) -- a suite of 24 datasets derived from a baseline C-SLAM front-end and real-world LiDAR data. Data DOI: https://doi.org/10.1184/R1/29652158
Daniel McGann, Easton R. Potokar, Michael Kaess
The authors are with the Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA.