Collision Avoidance

Photo of supertanker AbQaiq from Wikipedia

In all the hoopla about self-driving cars, I am reminded about collision avoidance systems for ships.

When I was in my teens, Sperry Rand was developing what would become the world’s first computerized collision avoidance system for ships. The prototype system went through its initial shakeout testing on the tanker where my father was Alternate Master. (That means he was Captain whenever the usual captain was away, and dropped down one rank when the usual captain was there.)

The Captain was not terribly excited about this. He had a couple of Sperry Rand engineers on board, an extra piece of equipment on the bridge that seemed a lot like another radar screen and a computer about the size of a big filing cabinet up there too. The new system couldn’t be relied upon yet. The Captain had a ship to sail. I suppose all this was a bother.

My father had always been interested in electronics, so he was fascinated. Whenever he had a bit of time to spare, he spent it with the Sperry Rand engineers.

I got to see, touch and get an introduction to the system once. The user interface seemed at first glance like just an echo of the radar system near waist height and at a good viewing angle for the operator standing behind it. Whatever was on the radar screen was also on the CAS screen, a big black circle with green on it wherever the radar saw something such as a shoreline or other vessels. It showed the sweep of the radar signal just like the radar system did.

It had a few controls a standard radar screen didn’t have, such as a small joystick. I could use the joystick to move a targeting window over something and press a button to tell the CAS I wanted to it to track that item. The system needed an operator who knew how to distinguish a boat from an island. Tracking an island wouldn’t be very useful. The CAS could track several targets, but not an infinite number, so in a busy area it was important to choose targets wisely. If memory serves, its limit was 20.

My father showed me how he could change settings in the CAS to say what timeframe he wanted it to predict, such as 10 minutes or 20 minutes, and how close the CAS should regard as too close. Then the CAS drew dotted lines from each target showing their predicted path over the timeframe. It showed the predicted path for my father’s ship over that timeframe too, and the “too close” buffer zone around his ship. If any of the targets were predicted to get too close, the system would show an alarm status. It was easy to see on the screen exactly which other vessels would get too close, where that would happen, and about how far into the predicted time window.

This may sound crude now, but it was a game changer for its era.

Once, my father had to navigate through a fleet of Soviet trawlers. During the Cold War, such fishing vessels were often spy ships in disguise. They played a game of swarming around American merchant vessels. A collision would be news that could be manipulated to make the USA look bad, and in a fleet of small vessels it was extremely difficult to avoid a collision. Ships are lumbering beasts. They can’t turn on a dime.

The CAS couldn’t track enough targets to keep track of the entire trawler fleet, but it could track most of them. My father used it to find the one path he could take to get through the fleet without hitting any of them. He said without the CAS, he couldn’t have done it.

Sperry Rand spent many long months ironing out the kinks in their CAS. IBM got wind of what they were doing and rushed to create a competing system. Starting so far behind, they couldn’t refine theirs as much. When the two systems hit the market, Sperry Rand’s was definitely best.

Gulf Oil, for which my father sailed, started installing the Sperry Rand CAS on its tanker ships.

As a result, one of their ships came frightfully close to a collision off the East Coast that would have caused a terrible oil spill.

Why? How could installing a wonderful new collision avoidance system result in a close call with disaster?

A bridge officer got the mistaken notion that the CAS could stand his watch for him.

Remember what I said about how the CAS needed an operator who chose targets wisely? It needed an operator who paid attention, dropping targets that changed course to move away and adding targets that came in near the edge of the screen looking like they might come closer. It wasn’t a system anyone could set and forget, but that’s what the officer did.

Gulf Oil paused its rollout of the CAS across its fleet. After considering the incident, Gulf decided officers needed more training whenever the CAS was installed on a ship. They needed to be drilled about not only what it could do, but also what it couldn’t do, and how to use it to stand watch better instead of thinking it could take over watch duties.

Gulf also decided to refocus the initial rollout on supertankers. The stakes were highest there and a supertanker took so long to make a turn that it made a ship of ordinary size look nimble. My father became a full captain around that time and then spent at least a year going from one supertanker to another, training officers to use the CAS properly.

Computers are much more powerful now than they were then. They can process more information, do more target selection for themselves, keep up with the faster pace of automotive traffic… but the computers for self-driving cars are nascent systems. Maturity is still ahead for them.

Personally, I wouldn’t be comfortable letting a self-driving car “stand my watch” unattended yet. As technology, it is a teenager. I’d prefer to see it grow up before I let it stand my watch.