AI Cut Bridge Near-Misses by Almost 20×: Is Machine Vision Ready to Become a Second Officer?
🔔 Subscribe to ShipUniverse Weekly →
The Test Before AI Joins the Bridge Team
A one-vessel pilot produced an extraordinary safety signal. Independent trials now show maritime computer vision detecting nearly every relevant target. The harder test is deciding how much responsibility those numbers actually justify.
Acheon Akti has tested an AI-enabled video safety system aboard one managed vessel. The reported bridge result is difficult to ignore: near-miss events fell by almost 20-fold, while recorded deviation time dropped from 8,839 minutes to 2,730.
The technology watches through onboard cameras, detects configured hazards or deviations, and alerts people while there is still time to intervene. The video can be processed locally, so the safety loop does not depend on a continuous satellite connection.
But watching continuously is only one part of bridge watchkeeping. A qualified officer also interprets intent, hears sound signals, communicates, applies COLREGs, understands the passage plan, judges uncertainty, challenges other people and takes responsibility for the vessel. That is the gap this report tests.
Translate the headline into what the data actually says
The monitored vessel spent 6,109 fewer recorded minutes in the conditions classified by the system as deviations.
A true twenty-fold reduction would leave only 5% of the original event rate. The public pilot description says “almost” twenty-fold, so 95% should be treated as an approximate interpretation.
Three datasets answer three different questions
Does behavior change?
One-vessel operational pilot. Strong signal for procedure monitoring and intervention, but the public data is too limited to separate AI effect from training, crew adaptation or voyage differences.
Can the machine see?
Independent live-vessel assessment measured perception performance against radar, AIS, visual observations and recorded evidence across an 828-nautical-mile voyage.
Does performance persist?
The 139-vessel dataset covers more than 10.8 million nautical miles and compares safety performance through each ship's first year of AI-assisted watchkeeping.
The visual perception layer is already performing at serious numbers
Mediterranean reality test
A feeder containership sailed Gioia Tauro to Marsaxlokk via Bar. The route included the Strait of Messina, port approaches, anchorage traffic and open water. Day and thermal cameras were assessed beside radar, AIS and human visual watchkeeping.
A second officer is much more than another pair of eyes
The title becomes useful when the job is broken into functions. Machine vision is becoming strong in some of them and remains fundamentally incomplete in others.
The strongest bridge team may be machine persistence plus human judgment
Where machine vision has an advantage
Where the officer still has the advantage
Six things have to be solved before the word “officer” becomes serious
Missed critical target
Average recall matters less if the rare missed object is an unlit fishing boat directly ahead.
Alarm fatigue
Persistent false or low-value alerts can train officers to ignore the system precisely when one warning matters.
Dirty or blocked optics
Salt, rain, spray, glare, icing, cargo stacks and physical obstruction can degrade a camera without disabling the rest of the bridge.
Unknown operating envelope
Performance measured in one route, weather range or traffic profile may not transfer unchanged to every ocean and vessel type.
Behavior changes because it is watched
A pilot crew may improve because procedures receive more attention during deployment, training and observation, not only because the algorithm intervenes.
Automation dependence
A successful digital lookout can create a new risk if the bridge gradually stops maintaining the independent human cross-check the system was meant to support.
The regulations currently treat AI as an additional means, not a certified deck officer
Lookout is multi-sensory
The vessel must maintain a proper lookout by sight and hearing and use all appropriate available means to appraise the situation and collision risk.
Watchkeeping carries responsibility
Navigational-watch arrangements must maintain a safe continuous watch, with the officer in charge responsible for safe navigation during the watch.
Autonomy still retains accountability
IMO's new non-mandatory autonomous-shipping framework emphasizes risk assessment, system design and human oversight, with the master retaining overall responsibility.
The next useful benchmark is not another dramatic percentage
Before machine vision is given greater operational authority, fleets need repeatable performance metrics across the conditions in which bridge officers actually work.
AI Watchkeeper Evidence Gate
Load the reported single-vessel pilot, the Lloyd's Register perception trial, the 139-vessel safety study or your own data. The tool separates perception quality, safety outcomes and evidence maturity instead of turning one percentage into a claim that AI can replace an officer.
How much bridge responsibility does the evidence justify?
Unknown data can be left blank. The model scores only published or entered evidence and always keeps legal command and certified watchkeeping outside the AI role.
The pilot supports continuous monitoring and earlier intervention, but public evidence is insufficient to establish full navigational-watch capability.
Current regulations and the evidence entered here support an assistive safety role while navigational responsibility remains with qualified humans.
We welcome your feedback, suggestions, corrections, and ideas for enhancements.
Please click here to get in touch