AquaBEV: Monocular Underwater BEV Occupancy with 3D Sonar Supervision
Authors: Trung Tien Dong, Shengji Jin, Chen Chen, Yi Sheng, Xiaomin Lin
Organizations: ERA Lab, University of South Florida, Tampa, Florida, USA. · YES Lab, University of South Florida, Tampa, Florida, USA. · University of Central Florida, Orlando, Florida, USA.
Autonomous underwater robots are widely used for exploration, monitoring, and inspection, where safe navigation depends on understanding the surrounding free and occupied space. Bird's eye view (BEV) occupancy provides such a representation, but predicting it from a single underwater RGB image is difficult due to limited, unreliable geometric cues from appearance alone. 3D imaging sonar offers complementary geometric measurements to supervise this task. We introduce AquaBEV, a monocular underwater occupancy model that predicts local BEV occupancy from a single RGB image, using paired 3D imaging sonar as geometric supervision during training. AquaBEV maps visual features into a calibration free polar representation and applies causal decoding along the range dimension before reconstructing the prediction in Cartesian BEV coordinates. A controlled underwater occupancy benchmark was established, adapting representative occupancy methods to the same RGB to sonar task under a unified protocol. AquaBEV achieves 31.4 Visible IoU and 38.6 Observed IoU, 4.0% and 4.3% relative improvements over the strongest transferred baseline.
Figures & tables
Figure 1: A single RGB image is used to predict local BEV occupancy.
Figure 2: Overview of AquaBEV. AquaBEV encodes a single RGB image into a learned polar representation ( θ , r), applies causal reasoning along the range dimension, and differentiably transforms the resulting polar occupancy prediction into Cartesian BEV coordinates (x, y).
Method
Design
Occupancy
Range
Geometry
Representation
Geometry
Visible ↑
Observed ↑
Near ↑
Mid ↑
Far ↑
R Macro ↑
bF1@2 ↑
Chamfer ↓
SurroundOcc [ 27 ]
Dense voxel
Projection based
29.8
34.9
38.9
40.6
18.9
32.8
48.4
0.355
BEVFormer [ 19 ]
BEV queries
Projection based
28.7
34.6
39.7
39.5
18.4
32.6
45.1
0.398
VoxFormer [ 18 ]
Sparse voxel
Projection based
29.2
35.7
38.8
39.6
19.7
32.7
47.0
0.371
MonoScene [ 3 ]
Dense voxel
Projection based
29.8
35.8
40.8
39.7
20.3
33.6
45.4
0.381
TPVFormer [ 13 ]
Tri perspective
Projection based
29.0
34.2
37.4
39.3
18.9
31.8
47.2
0.382
Table 1: Comparison of monocular underwater BEV occupancy methods on the proposed benchmark. All methods use the same RGB input, three state sonar supervision, data split, ConvNeXt B backbone, training schedule, and evaluation protocol ( BEST / SECOND BEST ).
Method
Vis. ↑
Obs. ↑
Near ↑
Mid ↑
Far ↑
R Macro ↑
bF1@2 ↑
Chamfer ↓
Cartesian decoder
30.8
37.4
42.4
42.0
20.7
35.0
47.7
0.351
Polar decoder
31.3
37.5
42.3
42.0
21.5
35.2
48.0
0.342
Polar + Survival supervision
31.3
38.5
42.6
42.0
21.7
35.7
45.8
0.344
Polar + Metric angular coupling
31.0
38.4
42.5
42.0
21.5
35.3
46.2
0.349
Polar + Full range context
30.9
38.6
41.1
42.0
22.1
35.0
44.1
0.360
AquaBEV
31.4
38.6
43.0
42.1
22.2
35.8
45.9
0.340
Table 2: Ablation of AquaBEV on occupancy and geometric quality. All results are averaged over three seeds.
Encoder
Params.
Val.
Vis. ↑
Near ↑
DINOv2 S
23.2M
30.6
28.2
39.0
ResNet 50
25.6M
33.9
32.7
45.3
ConvNeXt T
29.5M
33.5
30.9
42.6
ResNet 101
44.6M
33.6
30.1
42.1
ConvNeXt S
51.2M
33.5
29.1
40.1
ResNet 152
60.3M
33.8
31.3
43.9
Table 3: RGB encoder ablation. All encoders are evaluated using the same calibration free occupancy framework, training protocol, and held out test session. The results are from one seed.
Figure 3: Qualitative comparison of monocular underwater BEV occupancy.
Figure 4: Ablation comparison of Casual Range Decoder.
Safe underwater exploration requires a robot to understand where surrounding structures are located and which regions are available for motion. Existing vision-based underwater exploration systems commonly obtain this information indirectly by estimating monocular depth, unprojecting the geometry into 3D space, and accumulating it into a 2D bird's-eye-view occupancy map. This reliance on intermediate depth estimation is particularly problematic underwater, where scattering and wavelength-dependent attenuation degrade visual cues and limit the reliability of monocular depth estimates. We introduce AquaBEV-Nav, an underwater exploration framework that bypasses explicit monocular depth estimation through direct bird's-eye-view occupancy prediction. Built upon the CORAL hierarchical exploration framework, AquaBEV-Nav replaces its depth-based perception front end with AquaBEV. Given a single RGB frame, AquaBEV maps visual features into a learned polar representation, performs causal reasoning along the range dimension, and reconstructs local Cartesian occupancy without relying on intermediate depth prediction. The resulting occupancy map is accumulated into CORAL's persistent spatial memory, providing spatial context for VLM-based high-level planning and collision constraints for dynamics-aware local trajectory generation. Across ten simulated reef environments and six occupancy backbones evaluated under a single protocol, AquaBEV-Nav reaches 37.48 structure IoU and 53.2 target IoU, 88.95% closed-loop coverage with zero collisions.
Trung Tien Dong, Zhenqi Wu, Sahasra Kondapalli +3
1ERA Lab, University of South Florida, Florida, USA · University of Maryland- College Park, College Park, Maryland · 2YES Lab, University of South Florida, Florida, USA
Robust perception is essential for the deployment of autonomous underwater robots. However, optical cameras become unreliable under poor illumination and backscatter. Forward looking (2D) acoustic sensors remain effective under these conditions, but they measure range and bearing while leaving elevation unresolved, creating an ambiguity that prevents individual sonar returns from being localized in three dimensional (3D) space. This complicates the sensor use for 3D scene understanding and precise object detection. We introduce uScenes, a multimodal underwater dataset containing synchronized 3D multibeam sonar point clouds and RGB imagery. The dataset contains 110 scenes and 95,834 synchronized observation, representing 277.6 minutes of data collected across multiple field sessions. uScenes establishes a foundation for underwater sensor fusion, cross modal representation learning and 3D scene understanding. Code and datasets are given at https://github.com/era-research-lab/uScenes.
Trung Tien Dong, Zhenqi Wu, Aditya Penumarti +4
Embodied Robotics and Automation Lab, University of South Florida, Tampa, FL 33610, USA · Mechanical and Aerospace Engineering Department, University of Florida, Gainesville, FL 32611, USA · Naval Architecture and Ocean Engineering at the Seoul National University, Seoul 08826, South Korea
Autonomous navigation in 3D is a fundamental problem for autonomy. Despite major advancements in terrestrial and aerial settings due to improved range sensors including LiDAR, compact sensors with similar capabilities for underwater robots have only recently become available, in the form of 3D sonars. This paper introduces a novel underwater 3D navigation pipeline, called SHRUMS (Sensor Hallucination for Robust Underwater Motion planning with 3D Sonar). To the best of the authors' knowledge, SHRUMS is the first underwater autonomous navigation stack to integrate a 3D sonar and demonstrate safe operation in complex 3D environments without reliance on visual input, thereby remaining independent of visibility conditions. To accommodate the intricacies of the novel sensor data stream while achieving real-time locally optimal performance, SHRUMS introduces the concept of hallucinating sensor measurements from non-existent sensors with convenient arbitrary parameters, tailored to application specific requirements. The proposed concepts are validated in simulations informed with real 3D sonar sensor data, in-pool clean-water trials, and field deployments in real conditions with significantly degraded visibility conditions.
Susheel Vadakkekuruppath, Lauritz R. Fosso, Michail Chatzispyrou +4
Susheel Vadakkekuruppath and Jason M. O’Kane are with the Department of Computer Science and Engineering, Texas A&M University, USA · Herman B. Amundsen and Marios Xanthidis are with the Aquaculture Robotics and Automation, SINTEF Ocean