An autopilot at sea and an autopilot in a river are the same box doing two different jobs.
Offshore it holds a compass heading for hours while the watch does something else, and that is exactly what it was built for.
In a narrow channel the same function becomes a standing hazard, because the water is not going where the compass is pointing.
A channel bends. A heading does not.
Somewhere in every bend there is a moment when the two stop agreeing, and something has to intervene.
The question is what does the intervening.
Heading hold cannot follow a curve
The mode at the center of this does what its name says. It holds a bearing and corrects the rudder to keep holding it.
To turn, the operator either dials the setpoint around with a knob or takes the wheel, and those are not the same action at all.
Dialling the knob is comfortable and precise, which is why it gets used, but it delivers a slow steady turn at whatever rate the system chooses rather than the rate the bend requires.
Taking the wheel means disengaging, and disengaging a steering system is the moment when a vessel is momentarily under nobody’s control, since the machine has stopped and the human has not yet started.
In open water that gap costs nothing. There is nothing to hit for miles and a heading a few degrees off is corrected at leisure.
A bend removes the margin.
A tug, a deck barge and a bridge that stays up
The vessel was a 96 foot towing vessel with five aboard, a captain, a mate, two deckhands and an engineer.
Ahead of it was a deck barge just under 192 feet long and nearly 55 feet in the beam, loaded with precast concrete, being pushed rather than towed.
A pushed tow steers only from the stern, so the barge does not follow the boat so much as pivot around it, and the front of the unit swings wide of wherever the rudder puts the back.
The structure ahead was a railroad lift bridge across a tidal river, the kind that normally sits in the raised position so that boats pass freely underneath.
That detail decides who loses. A raised lift span is out of the way of the river and directly in the path of nothing at all, until something tall enough reaches it.
The waterway was never the one blocked.
Twelve degrees a minute, all the way in
The contact happened at 26 minutes and 41 seconds past four in the afternoon on June 15 of 2024, at about 5.5 miles an hour over the ground.
Roughly a minute and a quarter earlier the tow had begun turning to port at 12 degrees a minute, and investigators found that rate stayed constant until the moment of impact.
A constant rate is the tell. It is what a system holds, not what a person at a wheel produces, and nothing in the track shows anyone fighting the turn.
The mate said the autopilot switched without alarming and that he then went to hand steering, and investigators could not confirm any malfunction in the equipment.
The western span ended up displaced 6 feet 6 inches out of alignment, and the bill came to 15.8 million dollars.
Nobody reported it to anyone.
What the investigation could not settle
The probable cause is unusually candid about its own uncertainty, naming loss of control by the mate and then offering two possible reasons without choosing between them.
One is an error in switching out of autopilot. The other is impairment by fatigue. Both fit the recorded track equally well and the equipment itself was cleared.
That leaves the mate’s account as the only description of what happened in the wheelhouse, in the way that a machine locked in one position or a flaw found only afterward leaves one version standing.
What the report does state as a finding is narrower and more useful, which is that operators need to know every mode of the system well enough to leave it quickly.
The full sequence is laid out in the marine investigation report including the turn rate and the timing.
Training is the stated issue.
Two days of a bridge nobody checked
The part that should worry a regulator is not the steering. It is the silence afterward.
Neither the crew nor the company’s port captain contacted the Coast Guard or the railroad, although a vessel involved in a bridge strike is required to report it immediately.
So the damage sat undiscovered until June 17, when a train crew arrived, tried to lower the span for the crossing and found that it would not come down.
The span stayed out of service until November, running at reduced speed under manual control, and did not return to full remote operation until the following spring, which the trade coverage set out.
The lift span was hit by a barge doing walking pace.
The railroad found out from a stopped train.
Read the whole thing?
Get the week's signal, not the noise
Our sharpest reporting on energy, climate and nature — free, once a week.
