I understand that keeping on a desired course is not a simple task and requires prediction and understanding of the surrounding waters... But you can say the same thing about crosswinds and thermal layers and pockets of turbulence in the air over a runway, and autopilots land 747s every day. In a far more computationally intensive and less recoverable domain, with more complex system dynamics and controls even discounting the extra dimension, plus sensing that's both more limited and harder to interpret. If modern controls theory can land an aircraft, I'd expect it to be able to run a ship along a trajectory like it's nailed to a rail in space-time.

TL/DR: I have to boil this down to a gut feel, but I'd rather fly a known takeoff/landing profile at any airport, than take a n-thousand ton vessel through a port departure/arrival, simply because the overall situation in the air is so much simpler.

Entering/leaving any decent sized port that handles larger vessels is a much more intense collision avoidance situation than any airspace management situation I can imagine.

Reducing the reaction space from three to two dimensions and extending the reaction time from control input -> output by a factor of 100 to 1000 is part of it.

The other complication is having a magnitude or more of different classes of "threats", from unexpected solid objects ( chart errors or new items unreported), other vessels both larger and smaller with faster or slower response times of their own, errors in SA... eyes and professional judgment help immensely. Thus the focus on keeping the OOD watching the whole SA.

My sub had at least a dozen people involved in building the "picture" around the boat, visual, sonar, radar... plus two or three of us collating that into displays and reports the OOD could see immediately.

I'm not advocating full autopilot; I agree that safe trajectories can be harder to plan in dense 2d than in sparse 3d, and that military ships in particular often have high-level objectives that would be extremely difficult to formalize for input into a planner.

Rather, I'm doubting that going in a straight line at a target speed is correctly handled with a three-person steering wheel. My understanding is that the current OOD already has a trajectory planned through the workspace coordinate system (space, time) and gives orders in workspace coordinates ("go this fast") that are then translated into low-level control outputs ("set the rudder to X"). It sounds like in some cases the OOD even gives direct control outputs - "Prepare for X thrust in Y minutes". Certainly these low-level controllers can't be a good thing for the OOD to be spending cognitive resources on, and I'd be astounded if we can't replace that with a computer that translates directly from that trajectory to the control outputs.

That's why I used airplanes as my example, rather than, say, autonomous vehicles. A commercial airliner's path through the air to a safe landing is basically hard-coded. The task of the airplane's autopilot, then, is simply to follow that trajectory - and I'd guess that it's far more difficult for an airliner to follow a workspace trajectory than it is for a ship to do so.

Really, the hard part is going to be getting the trajectory out of the OOD's head. And even then, it'd be so much easier if the helmsman could execute "go this way" by punching "this way" into a computer instead of staring at a compass and directly and continuously controlling the rudder.

Well said, that gives me a perfect frame to say that the inherent complication in handling the ship's trajectory is how rapidly the trajectory may need to completely change, contrasted against the reaction time of the vessel itself.

The OOD is estimating a best trajectory given an analysis of the navigation constraints and the behavior of every possible contact in the vicinity. Orders at every level of detail are possible at any time... from 'resume best course for point "X-Ray"' to 'allaheadflankfullrightrudderbraceforimpactport!'. That is the point of resistance to most levels of automation past 'ship's wheel mirrors rudder angle' and 'engine order repeater indicates desired engine rpm'.

Aircraft, even big ones, are still far more maneuverable than large navy vessels. That extra Z dimension gives you more space to work through problems, even if it adds complexity.