Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: I'm Rosa.
Dev: And I'm Dev, here with Taro.
Rosa: We're looking at the paper: "Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin".
Paper summary: Rosa: So we're looking at this paper called "Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin." Basically, they’re tackling a big problem: testing autonomous ships is super expensive and risky on real vessels.
Dev: Right. The core idea here is that instead of just relying on simulations, which can miss real-world dynamics, they've built an actual setup using small model ships and robotic vessels in a lab basin to get early validation done safely.
Taro: I’m wondering if you can really test the complex stuff autonomy algorithms need when the ship runs into unexpected things outside of a controlled lab environment.
Rosa: That’s exactly what they're trying to address, Taro. The paper claims they developed an inhouse testbed that bridges that gap between digital simulations and physical testing for developing and validating MASS algorithms in a digital-physical laboratory <ref:2505.06787#pg1>
Dev: It sounds like the thesis is about creating this unified setup, which combines low- to high-fidelity simulation environments with physical models and a unified control interface for researchers and industry collaborators.
Taro: So, they have these different simulation levels, right? What’s the difference between what you can test in that low-fidelity model versus the high-fidelity one?
Rosa: Well, they have three complementary simulation environments. First, there's a low-fidelity model that uses simplified hydrodynamics to rapidly test software interfaces and high-level control logic. Then, they have a high-fidelity potential-flow based numerical model called mcsimpy2 that replicates the vessel dynamics by combining rigid body dynamics with frequency dependent hydrodynamics and wave excitation models.
Dev: That high fidelity level sounds crucial because it’s specifically targeted for developing and tuning the guidance navigation and control algorithms. How does that relate to what you were saying about testing outside the lab?
Taro: It means they can develop a solid controller in simulation, but then use the physical models to check if that controller actually works when you introduce real-world things like waves or unexpected motion on the physical fleet <ref:2505.06787#pg3>
Paper summary: Rosa: Exactly. And they also have a three dee simulation environment called Stonefish, which is used as a sandbox for image processing because it supports a wide range of sensor types like cameras and sonars <ref:2505.06787#pg3>.
Dev: I’m interested in the software side too. What kind of control stack are they running on for all these vessels? Is it consistent across the fleet?
Rosa: They've standardized everything on a shared software stack written in Python and C++. Robot Operating System two or ROS two is used as the middleware to handle communication between different components and external programs <ref:2505.06787#pg3>.
Taro: And what’s actually in that control stack? Is it just basic movement commands, or are they doing something more complex for autonomy?
Dev: The default modules include a thrust allocation module to distribute forces to thrusters, a proportional integral feedback velocity tracking controller, and a third order reference filter integrated with a PID dynamic positioning controller. Plus, they use an extended Kalman filter from the robot localization package.
Rosa: That sounds like a pretty comprehensive set of tools for navigation and control in real time. It’s all tied together through ROS two <ref:2505.06787#pg3>
Taro: So, if we look at the results they showed, the simulation achieved a position root mean square error of zero point sixty-two meters and a yaw root mean square error of four point thirty-nine degrees <ref:2505.06787#pg3>
Dev: And for comparison, the physical vessel showed an error of zero point one eight meters, four point two nine five degrees, and zero... wait, you stopped there. The physical results seem to be significantly better than the simulation <ref:2505.06787#pg3>
Rosa: That difference highlights why they need this testbed—it shows the simulation is a good starting point for development but doesn't capture everything, and the physical system gives them a more realistic benchmark. It also mentions that all hydrodynamic data is released publicly to support wider adoption of this testing approach <ref:2505.06787#pg3>
Taro: What does this actually mean for someone who isn't building MASS systems? It means they can get access to a standardized, open software-hardware stack that acts as a benchmark for autonomy research.
Paper summary: Dev: It’s about making the testing process more accessible by providing this modular infrastructure where researchers can plug in different components and see how they interact without having to build everything from scratch for every experiment.
Rosa: So, when we look at the overall picture of this paper, "Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin," it’s about creating a streamlined way to validate complex ship autonomy algorithms by integrating simulation environments with physical models and sensors into one accessible platform.
Dev: And ultimately, the authors are pushing for an open modular software hardware stack that includes a hydrodynamic model library, which they hope will transform the Marine Cybernetics Laboratory basin into a benchmark testbed for autonomy research <ref:2505.06787#pg3>
Taro: What about where this work stops? They mention the limitations are around the cost and cumbersomeness of designing and developing a model vessel for every single test, which is still something researchers struggle with <ref:2505.06787#pg1>
Rosa: So, while the testbed solves the access problem for early validation, they admit that building a physical model just for one specific test case can still be costly and cumbersome <ref:2505.06787#pg1>
Dev: That makes sense from a real-world deployment standpoint. The testbed helps with development but doesn't necessarily replace the need for those massive, full-scale tests later on when you’re dealing with actual operational costs and safety concerns <ref:2505.06787#pg1>
Taro: It seems like the implication is that this testbed is essential for moving from concept to a more robust system, allowing them to safely iterate on the autonomy algorithms before they move onto those expensive real-world trials <ref:2505.06787#pg2>
Rosa: So, in conclusion, "Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin" presents a concrete solution to testing autonomous ship algorithms by combining multiple simulation levels with physical vessels and sensors into one standardized environment <ref:2505.06787#pg1>
Dev: It’s about providing that open, modular software hardware stack that includes the hydrodynamic model library, which they hope will support wider adoption of autonomous system testing in shipping operations <ref:2505.06787#pg3>
Conclusion: Rosa: So, we've been looking at this paper about a digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin.
Dev: Basically, they built a way to test those tricky ship control algorithms safely outside of expensive real-world testing environments.
Rosa: Right. They took models and robotic vessels and put them in a lab basin to validate the software before you risk anything on an actual ship.
Dev: The authors are focusing on creating this unified setup, blending digital simulations with physical testing for developing autonomous ship systems.
Rosa: It seems they're pushing for an open stack of software and hardware that researchers can use to plug in different components easily.
Dev: That’s the big takeaway: a standardized platform that includes hydrodynamic data and simulation models, which should make testing autonomy more accessible.
Rosa: So, it’s about giving researchers a concrete way to iterate on their ideas without needing a massive shipyard for every little experiment.
Dev: The authors are hoping this testbed becomes the go-to benchmark for autonomous system testing in shipping operations globally.
Rosa: It means we can get a better look at how these algorithms actually perform when the world throws unexpected stuff at them, even if it’s just in a controlled setting.
Dev: That opens up possibilities for developing more robust and reliable systems before they ever hit the open sea.
Rosa: But they also admitted that building a physical model for every single test case can still be costly and take a lot of effort to set up.
Emir Cem Gezer, Mael Korentin Ivan Moreau, Anders Sandneseng Høgden, Dong Trong Nguyen, Roger Skjetne, Asgeir Sørensen
Norwegian University of Science and Technology - NTNU
cs.RO, cs.SY, eess.SY
Submitted: 2025-05-10
Updated: 2026-11-17
License: http://creativecommons.org/licenses/by-sa/4.0/
Importance score: 79/100
The gist: The algorithms developed for Maritime Autonomous Surface Ships (MASS) are often challenging to test on actual vessels due to high operational costs and safety considerations Utilizing small-scale…
Key concepts
- Testbed Infrastructure
- This is the integrated setup combining digital simulations with a physical testing environment. It allows for a progressive workflow where algorithms are developed and validated in a controlled setting, bridging the gap between virtual testing and real-world application.
- Low-fidelity Model
- This simplified model uses basic hydrodynamics to approximate vessel movement. It is used early on to rapidly test software interfaces and high-level control logic without needing complex calculations, making initial development faster.
- High-fidelity Model
- This numerical model uses potential flow theory to accurately replicate the dynamics of the ship, including rigid body motion and wave effects. This level of detail is crucial for tuning advanced guidance, navigation, and control (GNC) algorithms.
- ROS 2 Middleware
- Robot Operating System 2 acts as the central communication layer for all vessels. It allows different software components—like control modules or sensor processing—to talk to each other efficiently and seamlessly, ensuring the entire system works together effectively.
Terminology
Summary
The algorithms developed for Maritime Autonomous Surface Ships (MASS) are often challenging to test on actual vessels due to high operational costs and safety considerations Utilizing small-scale model ships and robotic vessels in conjunction with a laboratory basin provides an accessible testing environment for the early stages of validation processes To address these challenges and enable streamlined testing, we have developed an inhouse testbed that facilitates the development, testing, verification, and validation of MASS algorithms in a digital-physical laboratory
Testbed Infrastructure
The infrastructure integrates digital simulations with a physical testing environment to create a seamless, progressive workflow for developing and validating MASS algorithms At the core of this setup is the C/S fleet (see Table 1) The multi-layered infrastructure integrates digital simulations with a physical testing environment to create a seamless, progressive workflow for developing and validating MASS algorithms This setup includes low- to high-fidelity simulation environments and a streamlined interface for early-stage development and testing, with all hydrodynamic data released publicly to support wider adoption
Simulation Environments
The testbed combines simulation environments, hydrodynamic datasets, visualization tools, physical models, and a unified control interface intended for academic researchers and industry collaborators The infrastructure includes three complementary simulation environments tailored for different stages of the development and verification process
-
Low-fidelity model approximates vessel dynamics using simplified hydrodynamics following the definitions in (Fossen, 2021) This environment is primarily used for rapid testing of software interfaces and high-level control logic
-
High-fidelity model potential-flow based numerical model mcsimpy2 replicates the dynamics of the vessels in the C/S fleet by combining rigid-body dynamics with frequency-dependent hydrodynamics and wave-excitation models This level of modeling deliberately targets sufficient fidelity required for guidance, navigation, and control (GNC) algorithm development and controller tuning
-
3D simulation with sensor models Stonefish is used for providing a sandbox for image processing as it supports a wide range of sensor types such as cameras, inertial measurement units, sonars, etc. p. 7
Where the paper says it
3D simulation with sensor models Stonefish (Cie´slak, 2019) is a high-performance and real-time capable robotics simulator written in C++. Here, it is used for providing a sandbox for image processing as it supports a wide range of sensor types (such as cameras, inertial measurement units, sonars, etc.
Control and Estimation Software
All the vessels in the C/S fleet operate on a shared software stack which is written in Python and C++ programming languages Robot Operating System 2 (ROS 2) serves as the middleware layer, enabling efficient inter-process communication and seamless integration with additional components and external programs The default control modules include a thrust allocation module for distribution of forces to the thrusters, a proportional-integral (PI) feedback velocity tracking controller, a third-order reference filter integrated with a PID dynamic positioning controller, and an extended Kalman filter (EKF) from the robot localization package
Validation and Testing
The testbed enables safe, repeatable, and cost-effective testing of autonomy algorithms while also supporting more advanced use cases such as remote control and hybrid testing scenarios Example experiments show that the simulation achieved a position RMSE of 0.062m, yaw RMSE of 0. p. 6 5 cm and peak periods Tp = 0. The results are summarized in Figure 4b, which shows the mean of the surge, sway,
and yaw loads measured in the mooring lines in the physical basin versus those predicted as mean wave loads in the simulation. Software
3. Core modules
Inertial
Data
Inertial Measurement
Unit Driver
Pose Data
Motion Capture
System Driver
Control
System
Desired
Pose
Thrust
Allocation
Velocity
controller
Observer
Extended Kalman Filter
Robot Localization package
Thruster
manager
Various actuator
drivers
Pose
controller
Desired
Velocity
Desired
Thrust Load
Desired
Thrust Load
Figure 5. However, designing and developing a model vessel for a single test can be
costly and cumbersome, and researchers often lack access to such infrastructure. To
address these challenges and enable streamlined testing, we have developed an inhouse testbed that facilitates the development, testing, verification, and validation of
MASS algorithms in a digital-physical laboratory.Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Digital-physical testbed for ship autonomy studies can be used to establish a full design and verification pipeline that starts from low-fidelity and moves up to high-fidelity simulation models of each vessel, thereby to model-scale testing of the vessel in the laboratory basin.
-
The testbed enables safe, repeatable, and cost-effective testing of autonomy algorithms while also supporting more advanced use cases such as remote control and hybrid testing scenarios.
-
Algorithms can be tested on a growing fleet of Cybership (C/S) vessels spanning rapid model-scale testing in various simulated environments on TRLs 2-4 and controlled basin testing on TRLs 3-5, before moving to higher-level validation with half- and full-scale platforms.
-
The system supports symbolic control for autonomous docking of a dynamic positioning vessel by combining a real-time symbolic controller for desired surge, sway, and yaw velocities with the low-level velocity feedback loop.
-
Algorithms can be run in real time to monitor maritime navigation specifications written in Signal Temporal Logic (STL) extended with probably approximately correct (PAC) bounded reachability predictions to compute robustness bounds accounting for model uncertainty.
Abstract
The algorithms developed for Maritime Autonomous Surface Ships (MASS) are often challenging to test on actual vessels due to high operational costs and safety considerations. Simulations offer a cost-effective alternative and eliminate risks, but they may not accurately represent real-world dynamics for the given tasks. Utilizing small-scale model ships and robotic vessels in conjunction with a laboratory basin provides an accessible testing environment for the early stages of validation processes. However, designing and developing a model vessel for a single test can be costly and cumbersome, and researchers often lack access to such infrastructure. To address these challenges and enable streamlined testing, we have developed an in-house testbed that facilitates the development, testing, verification, and validation of MASS algorithms in a digital-physical laboratory. This infrastructure includes a set of small-scale model vessels, a simulation environment for each vessel, a comprehensive testbed environment, and a digital twin in Unity for visualization and remote control. With this, we aim to establish a design and verification pipeline that progresses from low-fidelity simulations, through potential-flow-based simulations of each vessel, to model-scale testing in the laboratory basin. Further advancement allows moving towards half-scale validation with R/V milliAmpere1 and full-scale validation with R/V Gunnerus. In this work, we present the testbed and its components and demonstrate its use for early-stage development of ship autonomy guidance, navigation, and control (GNC) algorithms. Model-scale hydrodynamic scaling, thruster characterization, and uncertainty propagation remain limitations for later-stage validation.
Sources
Related papers
- Rethinking Actions: Continuity-Aware Priors for World-Model Planning
- MOSAIC: Modular Obstacle and Spatial-Aware Framework for Manipulation In Cluttered Scenes
- SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
- DexSeed: Expanding Human Hand-Object Interactions into Robot Demonstrations for Dexterous Policy Learning
- LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
- CIALLO: Learning Shared Intent Representations for Cross-Embodiment Manipulation