1 The Hong Kong University of Science and Technology (Guangzhou)
2 Boston University 3 Shanghai Jiao Tong University 4 Tencent
Unified ocean-environment data, canonical multi-agent marine tasks, and cross-backend evaluation from scalable quantitative benchmarking to underwater scene deployment.
Simulation is essential for marine robotics, yet existing platforms often trade off visual realism, experimental throughput, and ocean conditions derived from real data. We present OneOcean, a data-grounded simulation suite and benchmark that combines bathymetry, currents, and tides in a shared spatiotemporal dataset. Optional pollution fields support pollution-response evaluation. OneOcean defines ten tasks covering navigation, sensing, coordination, and pollution response, with teams of 2–10 vehicles. The simulator supports three to six Degrees of Freedom (DoF) and records standardized metrics and run configurations across tasks, difficulty tiers, and team sizes. Experiments across dataset variants and underwater scenes compare heuristic control, Behavior Cloning (BC), and Large Language Model (LLM)-assisted planning under ocean currents.
OneOcean overview.
A unified NetCDF environment dataset harmonizes GEBCO bathymetry, Copernicus / GOPAF currents, optional tide terms, and derived masks. This dataset supplies ocean conditions for the ten-task benchmark in both simulation backends.
The core benchmark backend runs deterministic benchmark episodes at scale. The external underwater scene backend runs the same tasks in packaged underwater scenes using OneOcean currents.
OneOcean combines bathymetry and ocean currents in a shared dataset, exports currents to two simulation backends, and evaluates controllers on ten tasks with common metrics. The dataset can be regenerated for a chosen region, time window, and resolution.
The data pipeline merges GEBCO bathymetry with Copernicus / GOPAF currents and optional tides into a combined NetCDF environment. Variant exports keep the schema fixed while changing coverage and resolution for tiny, scene, and public releases.
The core backend runs task and difficulty sweeps. The external backend runs tasks in packaged underwater scenes using exported OneOcean currents.
Tasks use three difficulty tiers and report success, efficiency, and task-specific progress. The simulator supports 3–6 DoF dynamics; the quantitative study uses the 6DoF setting. Run records include the configuration and dataset indices.
Go-to-Goal under Currents (G2G), Station Keeping (SK), Route Following (RF), and Depth Profile Tracking (DPT) test disturbance rejection and sustained tracking under data grounded drift.
Area Scan (AS) and Pipeline Inspection (PI) stress coverage, event localization, and terrain-aware operation under the same grounded flow conditions.
Formation Transit (FT), Surface Pollution Cleanup (SPC), Underwater Pollution Lift (UPL), and Fish Protection / Herding (FH) emphasize team-size scaling, constraint satisfaction, localization, containment, and cleanup.
The benchmark compares navigation, sensing, and coordination across task difficulties and simulation backends.
Heuristic denotes the deterministic task-specific baseline. Behavior Cloning (BC) is trained to imitate heuristic actions. LLM planners select source assignments or waypoints; all LLM runs use the same low-level controller and safety checks. Highlighted entries indicate the best value in each column.
Across ten tasks, heuristic success decreases from 84.6% at Tier 2 to 29.0% at Tier 3; BC decreases from 58.6% to 6.0%.
Both backends use currents exported from the same ocean dataset and share task definitions.
Navigation tests disturbance rejection, inspection tests coverage and detection, and pollution response tests assignment and coordination.
Macro results over the 10-task suite plus Tier 3 key-task metrics. Higher is better for success, coverage, inspection score, and cleanup rate; lower is better for the action-energy proxy E and collision ratio.
| Method | T2 Suite SR (%) | T2 E | T2 Coll (%) | T3 Suite SR (%) | T3 E | T3 Coll (%) | Go2Goal SR (%) | Route SR (%) | Station RMS | Scan cov. | Pipe score | Cleanup rate |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Heuristic | 84.6 | 1842.3 | 5.68 | 29.0 | 3352.4 | 8.19 | 74.0 | 54.0 | 1.07 | 0.89 | 0.78 | 0.56 |
| BC | 58.6 | 1271.2 | 9.72 | 6.0 | 1635.1 | 6.41 | 34.0 | 0.0 | 26.52 | 0.89 | 0.67 | 0.06 |
Success rates across difficulty tiers on representative tasks.
Surface cleanup scaling shows the trade-off between improved cleanup outcomes and higher coordination cost.
Tier 2 planning results under stronger currents. LLMs select source assignments or waypoints; all LLM runs use the same low-level controller and safety checks.
| Method | SR avg (%) | Cleanup SR (%) | Cleanup rate | Scan SR (%) | Scan cov. | Pipe SR (%) | Pipe score | T (s) | E |
|---|---|---|---|---|---|---|---|---|---|
| Heuristic | 65.8 | 60.0 | 0.87 | 77.5 | 0.80 | 60.0 | 0.69 | 776.8 | 9132.4 |
| BC | 45.0 | 10.0 | 0.32 | 67.5 | 0.79 | 57.5 | 0.61 | 921.6 | 6908.2 |
| LLM-ChatGLM3-6B | 55.0 | 60.0 | 0.87 | 77.5 | 0.82 | 27.5 | 0.54 | 860.3 | 9671.5 |
| LLM-Qwen2.5-7B | 47.5 | 57.5 | 0.81 | 57.5 | 0.78 | 27.5 | 0.57 | 1077.2 | 11703.8 |
| LLM-Qwen2-7B | 45.0 | 57.5 | 0.86 | 67.5 | 0.81 | 10.0 | 0.51 | 991.4 | 10422.1 |
| LLM-Mistral-7B | 42.5 | 55.0 | 0.82 | 55.0 | 0.78 | 17.5 | 0.52 | 1044.2 | 11288.9 |
| LLM-Llama3-8B | 44.2 | 67.5 | 0.84 | 47.5 | 0.77 | 17.5 | 0.59 | 1099.1 | 11614.2 |
| LLM-Llama2-7B | 30.8 | 55.0 | 0.87 | 27.5 | 0.75 | 10.0 | 0.46 | 1156.4 | 11654.3 |
Route-following results across four underwater scenes using currents exported from OneOcean.
| Scene | Tier 2 | Tier 3 | ||||
|---|---|---|---|---|---|---|
| SR (%) | Waypoints | CTE (m) | SR (%) | Waypoints | CTE (m) | |
| Dam | 0.0 | 1.0 | 3.06 | 0.0 | 0.0 | 0.87 |
| OpenWater | 20.0 | 3.1 | 2.50 | 0.0 | 0.0 | 1.91 |
| PierHarbor | 56.7 | 3.5 | 2.22 | 0.0 | 2.0 | 1.41 |
| SimpleUnderwater | 100.0 | 4.0 | 3.04 | 96.7 | 3.0 | 2.16 |
| Mean over scenes | 44.2 | 2.92 | 2.70 | 24.2 | 1.25 | 1.59 |
Task demonstrations from the core and underwater-scene backends.
Multi-agent fish herding under ocean currents.
Multi-agent formation transit under ocean currents.
Surface plume containment behavior in a packaged harbor scene with OneOcean current injection.
Plume-source localization in PierHarbor.
Formation transit in OpenWater.