Traceable Virtual Sea Trials in the Marine Robotics Unity Simulator for Manoeuvring Assessment of Unmanned Surface Vehicles

Technical Summary: Traceable Virtual Sea Trials in the Marine Robotics Unity Simulator

Problem Statement
Accurate identification of hydrodynamic derivatives is fundamental for the control and autonomous navigation of Unmanned Surface Vehicles (USVs). While physical sea trials (Turning Circle and Zig-Zag) are the standard for assessing manoeuvrability per International Maritime Organization (IMO) and International Towing Tank Conference (ITTC) guidelines, they are often constrained by high costs, safety risks, and environmental variability. Conversely, computational fluid dynamics (CFD) offers high fidelity but lacks the real-time capability required for autonomy experimentation.

Existing robotics-oriented simulators (e.g., ROS-based Gazebo, Unity-based MARUS) provide a practical middle ground but suffer from a critical limitation: they typically treat commanded control inputs as equivalent to realised actuation. This assumption fails to account for actuator saturation, execution delays, and physical constraints, particularly in differential-thrust USVs where no physical rudder exists. Consequently, simulator-based System Identification (SI) datasets often contain biases that compromise the reliability of hydrodynamic-derivative estimation and Digital Twin (DT) calibration.

Methodology
The paper presents an extension to the open-source Marine Robotics Unity Simulator (MARUS) to create a "Traceable Virtual Sea Trial" framework. The methodology integrates a hybrid control architecture where ROS handles high-level manoeuvre logic (TC and Zig-Zag sequences) while Unity manages physics integration, hydrodynamics, and actuation.

Key methodological innovations include:

1. Command–Execution Separation: The framework explicitly distinguishes between the ordered rudder-equivalent command ($\delta_{cmd}$) and the executed rudder-equivalent proxy ($\delta_{exec}$). For differential-thrust USVs, $\delta_{exec}$ is calculated in real-time from the port and starboard thrust imbalance ($T_{stbd} - T_{port}$), rather than assuming the commanded input is perfectly achieved.
2. Execution-Grounded Timing: Manoeuvre onset ($t_0$) and reference headings are defined based on the threshold crossing of the executed actuation signal ($\delta_{exec}$), rather than controller intent. This ensures all kinematic metrics are aligned with the physical reality of the vessel's response.
3. Standardised Metrics Extraction: The framework automates the extraction of IMO/ITTC-aligned metrics, including Advance, Transfer, and Tactical Diameter for Turning Circles (TC), and Overshoot Angles ($\alpha_1, \alpha_2$) and Peak Yaw Rates for Zig-Zag (ZZ) tests.
4. SI-Ready Data Conditioning: A post-processing pipeline validates internal consistency (comparing logged yaw rates from the physics engine against reconstructed rates from heading differentiation) and organizes data into SI-ready datasets containing dual actuation signals ($\delta_{cmd}, \delta_{exec}$) and vessel states.

Key Contributions
The paper outlines three primary contributions:

• Automated Framework: A standardized, automated workflow within MARUS for conducting virtual TC and ZZ sea trials that adheres to IMO/ITTC procedures.
• Actuation-Traceable Logging: A novel data structure for differential-thrust USVs that logs the mismatch between commanded and realised actuation, addressing a known failure mode in simulator-based SI where commanded inputs are incorrectly treated as achieved actuation.
• Auditable Post-Processing: A workflow that produces compliant, SI-focused datasets, enabling more reliable hydrodynamic-derivative estimation and Digital Twin calibration by preserving the traceability of physical actuation.

Results
Case studies were conducted on a simulated 300 kg USV using the proposed framework. The results demonstrated:

• Repeatability and Compliance: All manoeuvres satisfied IMO MSC.137(76) criteria. For TC tests, the normalised advance and tactical diameter showed minor asymmetries (3.9% and 4.6–4.7% differences between port and starboard), attributed to execution-level thrust imbalances rather than control errors.
• Zig-Zag Performance: Overshoot excesses for both $\pm10^\circ$ and $\pm20^\circ$ tests remained below $1^\circ$, well within IMO limits. Peak yaw rates ranged from 4.1 to 5.8 deg/s, scaling with manoeuvre magnitude.
• Traceability Validation: Analysis confirmed systematic deviations between $\delta_{cmd}$ and $\delta_{exec}$ during saturation and reversal phases. Ignoring these deviations would introduce bias into SI. Furthermore, the strong correlation ($\rho = 0.98$) between logged and reconstructed yaw rates validated the internal physical consistency of the simulator and data pipeline.

Significance
The paper claims that by grounding virtual sea trials in realised actuation rather than commanded inputs, the proposed framework generates datasets that are not only compliant with international standards but also physically consistent. This traceability is essential for overcoming the limitations of current simulator-based SI, where the assumption of perfect actuator response leads to biased hydrodynamic models. The framework supports the development of more reliable Digital Twins and autonomous control systems by ensuring that the underlying manoeuvring data accurately reflects the physical constraints and execution realities of the USV.