[論文レビュー] Marinarium: a New Arena to Bring Maritime Robotics Closer to Shore

Ignacio Torroba, D. Dorner+15 more|arXiv (Cornell University)|Feb 26, 2026

Underwater Vehicles and Communication Systems被引用数 0

ひとこと要約

The paper introduces Marinarium, a modular underwater research facility with a digital twin and space-robotics integration, and demonstrates data-driven system identification for underwater vehicles using Koopman EDMD with RBFs, among other open-research experiments.

ABSTRACT

This paper presents the Marinarium, a modular and stand-alone underwater research facility designed to provide a realistic testbed for maritime and space-analog robotic experimentation in a resource-efficient manner. The Marinarium combines a fully instrumented underwater and aerial operational volume, extendable via a retractable roof for real-weather conditions, a digital twin in the SMaRCSim simulator and tight integration with a space robotics laboratory. All of these result from design choices aimed at bridging simulation, laboratory validation, and field conditions. We compare the Marinarium to similar existing infrastructures and illustrate how its design enables a set of experiments in four open research areas within field robotics. First, we exploit high-fidelity dynamics data from the tank to demonstrate the potential of learning-based system identification approaches applied to underwater vehicles. We further highlight the versatility of the multi-domain operating volume via a rendezvous mission with a heterogeneous fleet of robots across underwater, surface, and air. We then illustrate how the presented digital twin can be utilized to reduce the reality gap in underwater simulation. Finally, we demonstrate the potential of underwater surrogates for spacecraft navigation validation by executing spatiotemporally identical inspection tasks on a planar space-robot emulator and a neutrally buoyant \gls{rov}. In this work, by sharing the insights obtained and rationale behind the design and construction of the Marinarium, we hope to provide the field robotics research community with a blueprint for bridging the gap between controlled and real offshore and space robotics experimentation.

研究の動機と目的

• Provide a cost-effective, modular indoor facility for repeatable maritime robotics experiments close to shore.
• Enable multi-domain experimentation (underwater, surface, air) and integration with space robotics lab infrastructure.
• Bridge simulation and real-world testing via a digital twin and real-time ROS 2 integration.
• Explore data-driven system identification for underwater vehicles and Sim2Real bridging methods.
• Demonstrate space robotics validation using underwater surrogates.

提案手法

• Describe Marinarium design: 9x5x3 m water basin, retractable roof, dual MoCap systems, and ROS 2 integration.
• Develop a data-driven discrete-time dynamics model for underwater vehicles from high-fidelity tank data.
• Compare four dynamics models on tank data: Koopman EDMDc with RBFs, Double Integrator (DI), Fossen-based physics model, and Physics-informed NN (PINc).
• Use Gaussian RBFs to lift state to a higher-dimensional space and learn linear A,B in zk+1=Azk+Buk with a linear decoder C.
• Train and evaluate models on 12D state and 8 thruster inputs at 50 Hz, reporting endpoint RMSE for horizons H=1,10,100.
• Provide a reproducible EDMDc–RBF identification pipeline with k-means for RBF centers and ridge regression for A,B.]
• research_questions: ["Can a modular, shore-close facility like Marinarium accelerate underwater robotics research by enabling repeatable, realistic experiments?", "How well can data-driven models (especially Koopman EDMDc with RBFs) identify underwater vehicle dynamics compared to traditional physics-based or neural approaches?", "To what extent can a digital twin and multi-domain testing reduce the sim2real gap for underwater robotics?", "What is the feasibility of using underwater surrogates for space robotics validation?"]
• key_findings: ["The Koopman EDMDc with RBFs model achieves the best endpoint RMSE across 1, 10, and 100-step horizons in tank data.", "Baseline DI and the Fossen physics model perform worse than the Koopman approach, especially at longer horizons.", "PINc exhibits very high RMSE values, indicating poor performance on this dataset.", "The evaluated RMSEs are 0.0629 (1-step), 0.0831 (10-step), and 0.1859 (100-step) for Koopman EDMDc–RBF; 0.0784, 0.1088, 0.4683 for DI; 0.0765, 0.2122, 0.5788 for Fossen; and 8.7886, 9.1550, 9.1639 for PINc.", "The study validates a practical pipeline for data-driven dynamics identification in underwater robotics using high-fidelity tank data.", "The Marinarium enables a controlled environment to collect dense, low-noise datasets suitable for operator-theoretic system identification and sim2real bridging.

実験結果

リサーチクエスチョン

• RQ1 Can a modular, shore-close facility like Marinarium accelerate underwater robotics research by enabling repeatable, realistic experiments?
• RQ2 How well can data-driven models (especially Koopman EDMDc with RBFs) identify underwater vehicle dynamics compared to traditional physics-based or neural approaches?
• RQ3 To what extent can a digital twin and multi-domain testing reduce the sim2real gap for underwater robotics?
• RQ4 What is the feasibility of using underwater surrogates for space robotics validation?

主な発見

Model	1-step RMSE	10-step RMSE	100-step RMSE
Koopman (EDMDc–RBF)	0.0629	0.0831	0.1859
Double Integrator (DI)	0.0784	0.1088	0.4683
Fossen (BlueROV2)	0.0765	0.2122	0.5788
PINc (ResDNN)	8.7886	9.1550	9.1639

• The Koopman EDMDc with RBFs model achieves the best endpoint RMSE across 1, 10, and 100-step horizons in tank data.
• Baseline DI and the Fossen physics model perform worse than the Koopman approach, especially at longer horizons.
• PINc exhibits very high RMSE values, indicating poor performance on this dataset.
• The evaluated RMSEs are 0.0629 (1-step), 0.0831 (10-step), and 0.1859 (100-step) for Koopman EDMDc–RBF; 0.0784, 0.1088, 0.4683 for DI; 0.0765, 0.2122, 0.5788 for Fossen; and 8.7886, 9.1550, 9.1639 for PINc.
• The study validates a practical pipeline for data-driven dynamics identification in underwater robotics using high-fidelity tank data.
• The Marinarium enables a controlled environment to collect dense, low-noise datasets suitable for operator-theoretic system identification and sim2real bridging.