HoloOcean 2.0: Marine Robotics Simulator

• HoloOcean 2.0 is a high-fidelity marine robotics simulator that integrates advanced hydrodynamic modeling and photorealistic rendering using Unreal Engine 5.3.
• The simulator employs robust physics with Fossen 6-DOF equations and nonlinear damping, achieving simulated vehicle dynamics within 2% of real-world AUV data.
• Advanced sensor simulation—including ray-traced sonar and semantic cameras—and native ROS 2 connectivity support comprehensive testing of perception and autonomy algorithms.

HoloOcean 2.0 is a state-of-the-art, high-fidelity marine robotics simulator built on Unreal Engine 5.3, designed for the development and validation of autonomous underwater vehicles (AUVs) and marine robotic systems. It supports both perception and autonomy research, enabling Hardware-in-the-Loop (HIL), Software-in-the-Loop (SIL), and full autonomy stack validation with direct ROS 2 integration, advanced hydrodynamic modeling, high-performance sensor simulation, and photorealistic rendering (Romrell et al., 7 Oct 2025, Meyers et al., 10 Nov 2025, Aldhaheri et al., 8 Apr 2025).

1. System Architecture

HoloOcean 2.0 is organized around modular subsystems tightly integrated within Unreal Engine 5.3 and exposed to end-users via a Python/ROS 2 interface. The core components are:

• Rendering/World Manager: UE 5.3 provides Nanite Virtualized Geometry for high-detail meshes, Lumen global illumination (enabling dynamic underwater caustics and reflections), and World Partition for streaming large-scale environments.
• Physics & Dynamics Manager: Utilizes Unreal's built-in physics for collisions and constraints, with a custom Python/C++ implementation of Fossen-based 6-DOF hydrodynamics for underwater vehicles.
• Sensor Models: Includes cameras (RGB, depth), LiDAR, ray casting–based sonar (sidescan, forward-looking, bathymetric), and semantic labeling for vision/sonar ground truth.
• Middleware & Bridges: The Python client library runs within UE, serving as a bridge to external autonomy stacks, with full support for ROS 2 (topics, actions, services).

Simulation operates at a default tick of 30 Hz, with vehicle dynamics and physics substepping configurable up to 240 Hz for precise collision and constraint resolution (Meyers et al., 10 Nov 2025). Each iteration processes physics/rendering, computes forces and torques from hydrodynamic models, resolves collisions, samples simulated sensors, and conducts bidirectional communication with ROS 2 middleware.

Comparative architecture benchmarks demonstrate that HoloOcean 2.0's migration from UE 4.27 to UE 5.3 yields significant gains in environment detail, real-time lighting, and rendering performance, albeit with higher hardware requirements (Romrell et al., 7 Oct 2025, Aldhaheri et al., 8 Apr 2025).

2. Vehicle Dynamics and Hydrodynamic Modeling

The simulator implements Thor Fossen's standard 6-DOF rigid-body equations of motion for underwater vehicles:

$$M\dot{\nu} + C(\nu)\nu + D(\nu)\nu + g(\eta) = \tau$$

where:

• $\nu = [u, v, w, p, q, r]^T$: body-fixed velocity (linear + angular)
• $\eta = [x, y, z, \phi, \theta, \psi]^T$: inertial position + Euler angles
• $M$: combined rigid-body mass and added mass matrix
• $C(\nu)$: Coriolis and centripetal (including added mass)
• $D(\nu)$: linear + quadratic hydrodynamic damping
• $g(\eta)$: hydrostatic restoring contributions
• $\tau$: control inputs (fins, thrusters) (Romrell et al., 7 Oct 2025, Meyers et al., 10 Nov 2025, Aldhaheri et al., 8 Apr 2025)

Key features in version 2.0:

• Fully nonlinear quadratic damping
• Coupled rigid-body and added-mass Coriolis effects
• Arbitrary control surface placement and individual actuator dynamics ($$\dot\delta_i = \frac{1}{T_i}(\delta_{cmd,i}-\delta_i)$$)
• User-configurable parameters: environmental (density, currents), physical (mass, inertia), hydrodynamic (added mass, damping), hydrostatic (buoyancy, mass offsets), and control surfaces.

Vehicle and autopilot configurations are specified via Python or YAML scenario files, enabling programmatic scenario scripting and diagnostics. The Dynamics Manager synchronizes with the Unreal physics engine to apply calculated forces, and the validation process has confirmed simulated vehicle responses within 2% of vendor data for REMUS 100 AUV depth-hold and trim (Romrell et al., 7 Oct 2025, Meyers et al., 10 Nov 2025).

3. Sensor Modeling and Simulation

HoloOcean 2.0 supports a comprehensive range of marine sensors with advanced simulation fidelity:

• Sonar (Sidescan, Profiling, Imaging): The ray-traced approach eliminates legacy octree-based precomputation, enabling dynamic, memory-efficient, GPU-accelerated echo simulation via Unreal's RTX pipelines. For a single-beam echo sounder, the runtime reduced from 0.055 s/tick (octree-based) to 0.012 s/tick (ray casting), with all environment changes immediately reflected in the sensor stream (Romrell et al., 7 Oct 2025).
• Semantic Sensors: All camera and sonar modalities provide per-pixel or per-return semantic labeling via UE asset tags, supporting detailed ground-truth output for vision-related research.
• Other Sensors: Depth (pressure), IMU (6-axis), DVL (Doppler), magnetometer, and acoustic positioning. Sensor outputs include modeled noise processes (white Gaussian and Gauss–Markov), quantization, and bias drift, following domain-typical specifications (Meyers et al., 10 Nov 2025, Aldhaheri et al., 8 Apr 2025).
• Calibration Routines: Built-in batch calibration scripts allow users to estimate sensor bias and scale factors from controlled motion primitives.

The acoustic and optical image-formation models incorporate light attenuation, backscatter, and environmental conditions (turbidity, caustics), with pixel intensities computed from wavelength-dependent exponential attenuation:

$I(z) = I_0 \cdot e^{-kz}$

where $k$ is the attenuation coefficient, $z$ is the depth coordinate (Romrell et al., 7 Oct 2025).

4. ROS 2 Integration and Autonomy Stack Connectivity

Native ROS 2 support is realized via a Python bridge that translates simulated actuator and sensor data to standard message types, enabling seamless use with external planners, controllers, and hardware:

• Published topics: Odometry, IMU, depth, altitude, magnetic field, and sonar, at rates up to 100 Hz, with custom message types for advanced sensors.
• Subscribed topics: Fin/thrust commands, setpoints for depth, heading, and general Wrench inputs.
• Services and actions: Simulation reset, pause/unpause, vehicle spawn, and trajectory following.
• Quality of Service (QoS): Tunable settings for sensor/command topic reliability and delivery depth.

A typical ROS 2 launch pipeline covers instantiation of the HoloOcean simulation, vehicle control, and online data visualization. Integration supports both software-in-the-loop and hardware-in-the-loop configurations, where the actual vehicle’s control stack interfaces directly with simulated sensors via ROS 2 middleware, enabling closed-loop experiments with real controllers (Romrell et al., 7 Oct 2025, Meyers et al., 10 Nov 2025, Aldhaheri et al., 8 Apr 2025).

5. Environmental and Scenario Modeling

Environmental realism is advanced via several UE-integrated techniques:

• Procedural Generation: Python-driven scripts automate terrain and asset creation, allowing randomized world generation for robust learning and evaluation.
• Bathymetry Import: GeoTIFF or XYZ bathymetric grids produce detailed seafloor models. Utilities such as HolooceanBathymetryImporter provide down-sampling and smoothing.
• Volumetric Effects: 3D vector fields model environmental currents, with per-tick velocity disturbances applied to vehicle dynamics.
• Wave Modeling: UE 5 water plugins enable both Gerstner (parametric) and FFT-based (spectral) wave fields. Vehicle hull contact points sample local wave state for buoyancy and hydrodynamic force computation.
• Material Effects: Depth-dependent postprocessing and volumetric particle systems approximate water turbidity and suspended sediment fields.

Environmental parameters (e.g., terrain complexity, asset density, water conditions) are fully scriptable, allowing batch domain randomization for sim-to-real transfer studies (Romrell et al., 7 Oct 2025, Aldhaheri et al., 8 Apr 2025).

6. Hardware-in-the-Loop, Software-in-the-Loop, and Validation

HIL and SIL setups exploit HoloOcean 2.0's real-time ROS 2 communication for end-to-end autonomy testing:

• HIL Workflow: Vehicle control computers (e.g., CougUV) interface with the simulator over Ethernet; sensor messages from HoloOcean feed the actual controller, which returns actuator commands to the simulation. Time synchronization (ROS 2 clock or NTP) and rigorous logging (100–200 Hz) are used for validation and analysis.
• SIL Workflow: Both the complete autonomy stack and the simulation run as separate containers or processes on the same host, supporting mission-level validation and regression testing (Meyers et al., 10 Nov 2025).
• Performance Metrics: Benchmarked scenarios have shown depth-hold RMS errors of 0.015 m (simulation) vs. 0.022 m (field); rise/settling times and heading-control metrics in simulation track closely with real AUV results. For a flat terrain, altitude RMSE is 0.05 m in simulation, 0.07 m in field. End-to-end control loop latency remains around 12 ms (std. 4 ms).
• Resource Usage: HIL simulation consumed 26.5% CPU and 8.21% RAM; field-deployed systems show comparable profiles (Meyers et al., 10 Nov 2025).

These results demonstrate centimeter-level simulation accuracy across multiple domains and close agreement between simulated and experimental vehicle dynamics, supporting high-fidelity development and pre-deployment validation.

7. Applications, Limitations, and Future Directions

HoloOcean 2.0 is applied to rapid mission prototyping (single or multi-agent), photorealistic visual training (e.g., underwater SLAM, segmentation), sim-to-real transfer, batch environment generation, and performance assessment of navigation/sensor pipelines (Romrell et al., 7 Oct 2025, Aldhaheri et al., 8 Apr 2025). Notable case studies—such as Actag visual-acoustic localization and natural language-driven control—have demonstrated successful transfer of algorithms trained in HoloOcean to physical AUVs with sub-decimeter error and limited performance degradation (Aldhaheri et al., 8 Apr 2025).

Current limitations include:

• High hardware requirements (30–50% above Gazebo-based simulators for equal scenarios)
• No explicit small-scale wave or variable free-surface modeling
• Uniform current fields; no spatial gradients or advanced wind/storm models
• Sonar returns use single-bounce propagation; multipath handled approximately
• No standardized multi-vehicle acoustic networking stack out-of-the-box
• Bridge and Docker image maintenance required for latest ROS 2 releases

Planned future enhancements address GPU-accelerated fluid fields, real-time parameter identification, expanded acoustic communication (e.g., modal beamforming and multi-path channel simulation), multi-agent benchmarking, and tighter digital twin integration for online mission supervision (Romrell et al., 7 Oct 2025, Aldhaheri et al., 8 Apr 2025, Meyers et al., 10 Nov 2025).

Feature	Status/Metric	Source
ROS 2 support	Native; bidirectional bridge for topics/actions	(Romrell et al., 7 Oct 2025, Meyers et al., 10 Nov 2025, Aldhaheri et al., 8 Apr 2025)
Physics model	Fossen 6-DOF, full nonlinear damping & Coriolis	(Romrell et al., 7 Oct 2025, Meyers et al., 10 Nov 2025)
Sensor fidelity	Ray-traced sonar/camera, parametric noise models	(Romrell et al., 7 Oct 2025, Aldhaheri et al., 8 Apr 2025)
HIL/SIL Test error	Depth RMS: 0.015 m sim, 0.022 m field	(Meyers et al., 10 Nov 2025)
Real-time rendering	30 FPS (UE 5.3 Nanite + Lumen, RTX 2080 Ti)	(Romrell et al., 7 Oct 2025)

HoloOcean 2.0 thus provides a reference platform for reproducible, high-fidelity marine robotics research, bridging the gap between simulation and field deployment through rigorous physical and sensor modeling, real-world connectivity, and extensibility.