Roberto Galeazzi, leader of the Control & Robotics group at the Technical University of Denmark, presented a vision that redefines the path towards autonomous shipping. The key, he argues, is not to build ships that replace sailors, but to create an alliance where Artificial Intelligence learns and embodies the quintessentially human art of good seamanship.
Is Autonomous Navigation a Solved Problem?
The central question is whether autonomous navigation is a solved problem. Despite a decade of progress, current systems excel only in simple, open-water scenarios. The true challenge lies in confined and congested waterways, where rules alone are insufficient. Most autonomous collision avoidance algorithms focus on a subset of the IMO COLREGs, particularly the steering and sailing rules (Part B). However, Rule 2 of the COLREGs establishes that strict rule-following must be tempered with good seamanship – the nuanced judgment, prudence, and best practices honed through experience. This quality has so far been the exclusive domain of human officers.
Codifying Good Seamanship: The Three Pillars of Machine Judgment
Galeazzi’s core research is the codification of good seamanship into machine-readable logic. For an AI to possess it, three capabilities are essential: dynamic risk evaluation (assessing collision and grounding threats in real-time), multi-step trajectory planning, and predictive scenario analysis. His team’s system integrates these by creating a comprehensive risk metric. This metric evaluates not just the risk from other ships, but also from the seabed (using electronic chart data), and even models the probability that a target ship might change speed. This continuous risk assessment feeds an advanced path-planning algorithm that prioritizes absolute safety before optimizing for voyage efficiency.
From Replacement to Alliance: The Generative AI Co-Pilot
This technical foundation leads to a crucial paradigm shift. The narrative that removing the human eliminates human error is flawed; it ignores the invaluable role of human expertise in preventing accidents. The next step is to infuse intelligent systems with this operational knowledge. Here, Generative AI and Large Language Models (LLMs) offer a revolutionary tool. A maritime-specialised AI could act as a collaborative co-pilot: analysing complex scenarios, engaging in dialogue with the officer on the bridge, and jointly deciding the safest action. Critically, through this dialogue, the AI would learn from the seasoned judgment of seafarers, continuously refining its own “good seamanship.”
The Reality Check: Why Off-The-Shelf AI Falls Short
However, a reality check is necessary. Tests of off-the-shelf LLMs (including GPT-4o) on 50 real navigation scenarios revealed severe shortcomings in their maritime reasoning. They frequently misinterpret rules or suggest unsafe maneuvers. This underscores a vital point: domain-specific development is non-negotiable. General-purpose AI cannot be trusted with critical navigation decisions; maritime intelligence must be built from the ground up.
Conclusion: The Way Forward
The safe operation of Maritime Autonomous Surface Ships (MASS) in complex environments hinges on a human-machine alliance. The goal is to plug the principles of good seamanship into robust autonomous systems, using generative AI not as an oracle, but as a framework to capture, scale, and apply hard-won human expertise. This partnership—not a replacement—promises to enhance situational awareness and safety beyond what either human or machine could achieve alone.
Source: Presentation by Prof. Roberto Galeazzi, Technical University of Denmark, January 2026. Analysis for MRANObservatory. https://electro.dtu.dk/