Autopilot response to waves

[RaymondR]

Originally Posted by rgleason

Well, Raymond (and Stu for that matter) let's just say we don't agree, but maybe we will live to find out.

Stu and I am actually on the same page.

The PID controller algorithm which, whilst pretty ancient, is still one of the most utilized mathematical model machine control methods. It came about when the US navy hired a scientist to improve autopilots. The scientist then observed how good helmsmen on ships applied steerage to control the course of the vessel and devised a mathematical model to emulate them.

Helmsmen on big ships don't have the problem of the ship being thrown about by large waves as do yachties so we need an add on to the model to compensate for this factor.

Stu and I are presenting our analysis of what is the factor we as helmsmen use to anticipate the effect of a large wave overtaking the vessel and we both propose that it is the magnitude of the acceleration we feel as the boat lifts which indicates the severity of the subsequent wave action and how much helm we should apply to counter the wave action.

In this we are emulating the scientist who devised the original PID control model in that we propose emulating the human based upon the same input the human is given by his or her senses.

[rgleason]

When a pedestrian steps off the curb and an autonomous vehicle swerves or stops is when these sensors we are talking about are working. The same thing can occur with waves, the software can be trained to ignore what is not relevant and react to what is relevant. It will happen first in boats like Don has sailed.

"One of the functions of the Expedition software is to provide the percentage actual speed vs polar data for each helmsman."

Thanks, opensource Tactics_pi should probably be extended to provide this feature, to make the "owners" and other bad helmsman feel better. Fortunately not all "owners" are bad helmsman (sounds like boat envy to me).

[rgleason]

Actually Tactics_pi by Tom and Dashboard_tactics by Petri both have Polar Performance

https://opencpn-manuals.github.io/pl...ar_performance

The manual is missing some images which need to catch up to a recent move from dokuwiki to ascii/antora but it is quite comprehensive. So you can rate your owner for free.

[AKA-None]

Originally Posted by RaymondR

Life gets even more complicated once one realizes that the vessels response is consequent upon two factors: the approaching wave, and it's motions in response to former wave trains. My recollection is that it is often the second or later waves in a big set which causes the largest course deviations.

"This is where I'm having trouble understanding how anything in an AP "system" can react to "random wave sets." Or, here's that word again, ANTICIPATE. That's why I steered by FEEL in anticipation of which way the boat would go. Very few of those experiences had ANY "regularity" to the wave sets. They felt almost always random."

Also what I recall Stu and the instinctive response when you feel the acceleration is to apply counter helm and is the reason I tend to feel the accelerometer triggering a rapid application of counter helm might be the the most practicable solution.

Can you define what you mean by feel?
Do you see something do you feel the rising boat, the pressure on the tiller? What?

[Stu Jackson]

Originally Posted by AKA-None

Can you define what you mean by feel?
Do you see something do you feel the rising boat, the pressure on the tiller? What?

Feel = sense, experience


rising boat, pressure on tiller, acceleration or deceleration


All of the above.


In one of my earlier posts, the one place I wasn't looking was backwards, or even sideways. That's what I mean by feel. I could almost have had my eyes closed.

[wingssail]

Originally Posted by AKA-None

Can you define what you mean by feel?
Do you see something do you feel the rising boat, the pressure on the tiller? What?

This is the key question. What is it that the helmsman "feels"? Understand that and learn to measure it and maybe you can get somewhere.

[Stu Jackson]

Originally Posted by wingssail

This is the key question. What is it that the helmsman "feels"? Understand that and learn to measure it and maybe you can get somewhere.

WS, I think that is exactly the heart of the issue: how does one get a machine to replicate the action of a human being?

The issue, at least to me and how I've tried to express them in this entire thread, is how to get a machine to deal with anticipation and feel.

The answer, so far, as I understand it, is that machine learning can be developed based on measurements (cameras determining wave height, frequency, direction, period, etc.) coupled with "knowledge" of boat behavior (which seems to have already happened with commercially available APs when they say their autopilots "learn" the boat).

My difficulty is in understanding just how those measurements can be used to deal with random non-periodic occurrence, which are exactly the issues one deals with in keeping course with following or quartering seas.

********************************



Separate comment for Rick @rgleason: Yes, autonomous automobiles can spot a pedestrian crossing a street. I personally don't understand how that can translate into controlling autopilots. Why? Because having a camera observe an obstruction is simple - it's either there or not; if it is there, braking or turning is the solution, depending on speed. Is it not?

If not, why not?

If so, how does this apply to autopilots in following seas?

Thx.

[goboatingnow]

There is no theoretical reason why vision systems that sense pedestrians couldn’t be used to provide an auto pilot with prediction capabilities. But the practical realities and complexities are beyond the computing power available on a boat.

Measuring boat movement is not prediction , a boat moves because the action that needs to be corrected is occurring right now

With quick computing and more importantly fast response steering systems a modern AP would most likely out perform most humans today just using non predictive inputs. This has been shown to be the case in single handed ocean racing.

It’s just such systems are far too expensive for leisure yachts.

True prediction has to use vision because there isn’t anything else.

To see this effect study helming in daylight and at night. Night helming is inevitably poorer even though all the same “ feelings “ except for sight are still present. Day time helming is better because good helmsmen “ read “ the seas and combined with experience ( or fore knowledge ) can react sometimes better.

[AKA-None]

Originally Posted by Stu Jackson

Feel = sense, experience


rising boat, pressure on tiller, acceleration or deceleration


All of the above.


In one of my earlier posts, the one place I wasn't looking was backwards, or even sideways. That's what I mean by feel. I could almost have had my eyes closed.

So all of those are measurable items then

[Stu Jackson]

Originally Posted by goboatingnow

1. There is no theoretical reason why vision systems that sense pedestrians couldn’t be used to provide an auto pilot with prediction capabilities. But the practical realities and complexities are beyond the computing power available on a boat.

2. Measuring boat movement is not prediction , a boat moves because the action that needs to be corrected is occurring right now

3. With quick computing and more importantly fast response steering systems a modern AP would most likely out perform most humans today just using non predictive inputs. This has been shown to be the case in single handed ocean racing.

4. It’s just such systems are far too expensive for leisure yachts.

5. True prediction has to use vision because there isn’t anything else.
To see this effect study helming in daylight and at night. Night helming is inevitably poorer even though all the same “ feelings “ except for sight are still present. Day time helming is better because good helmsmen “ read “ the seas and combined with experience ( or fore knowledge ) can react sometimes better.

I was thinking about approaching this, say, backwards. As in: What if cameras weren't really needed at all?

Reason is, I thought, that my earlier post mentioned that I could have done this with my eyes closed because I wasn't looking behind me. I was unaware of Item #5. For those of you who have been in heavy weather going downwind offshore, therefore both day and night, what have been your experiences? Not doubting gbn's #5, just trying to learn.

#s 1, 3 & 4 make sense.

But I don't understand the non-predictive part of #3. I thought the cameras, regardless of computing power and scale to our yachts seemed to be the barrier, not the fact that it couldn't be done.

OTOH, my "feel...no need to look" seems to support the non-predictive, so what help would the cameras provide if indeed the response by a helmsman is based on what is happening, rather than what is, possibly (i.e., cameras), about to happen?

Why would the cameras help?

Do good helmsmen actually look behind them?

[AKA-None]

Originally Posted by Stu Jackson

I was thinking about approaching this, say, backwards. As in: What if cameras weren't really needed at all?

Reason is, I thought, that my earlier post mentioned that I could have done this with my eyes closed because I wasn't looking behind me. I was unaware of Item #5. For those of you who have been in heavy weather going downwind offshore, therefore both day and night, what have been your experiences? Not doubting gbn's #5, just trying to learn.

#s 1, 3 & 4 make sense.

But I don't understand the non-predictive part of #3. I thought the cameras, regardless of computing power and scale to our yachts seemed to be the barrier, not the fact that it couldn't be done.

OTOH, my "feel...no need to look" seems to support the non-predictive, so what help would the cameras provide if indeed the response by a helmsman is based on what is happening, rather than what is, possibly (i.e., cameras), about to happen?

Why would the cameras help?

Do good helmsmen actually look behind them?

Well what do you do at night when you really can’t see far around you

[Stu Jackson]

Originally Posted by AKA-None

Well what do you do at night when you really can’t see far around you

Yes, that's another very good way to pose my question.


And relates to gbn's #5 in my last post.

[Rucksta]

Lots of comments about wave action being random.
Perhaps the perceived chaos is just a pattern that is more complex than the observer can interpret?

[goboatingnow]

Originally Posted by Stu Jackson

Yes, that's another very good way to pose my d canquestion.


And relates to gbn's #5 in my last post.

My point is predictive techniques on a yacht are based on observation ,and coupled with stored knowledge , prior experience and some intelligence.

One could add hearing , but my experience is it’s not a great predictor.

Prediction is about anticipating the future. It’s not based on “ feel” feel means sensing a boat physical effect however minute and coupled with experience and knowledge reacting to those sensory inputs. This is a reaction not a prediction . Any good software system with appropriate drive systems can actually do a better job then a human based on reactive inputs.

Hence for machine prediction have to add sight ( and maybe satellite telemetry ) in order to feed into the control system some future predictions

At night computer systems have the advantage of better visibility then humans.

The issue is cost , complexity and solving significant technical challenges. We are 10-20 years away from truely autonomous cars. We are possibly decades away from such technology being applied cheaply to yachts.

And yes helmsmen always look behind.

[RaymondR]

Originally Posted by goboatingnow

There is no theoretical reason why vision systems that sense pedestrians couldn’t be used to provide an auto pilot with prediction capabilities. But the practical realities and complexities are beyond the computing power available on a boat.

Measuring boat movement is not prediction , a boat moves because the action that needs to be corrected is occurring right now

With quick computing and more importantly fast response steering systems a modern AP would most likely out perform most humans today just using non predictive inputs. This has been shown to be the case in single handed ocean racing.

It’s just such systems are far too expensive for leisure yachts.

True prediction has to use vision because there isn’t anything else.

To see this effect study helming in daylight and at night. Night helming is inevitably poorer even though all the same “ feelings “ except for sight are still present. Day time helming is better because good helmsmen “ read “ the seas and combined with experience ( or fore knowledge ) can react sometimes better.

Remount the vision system from one of these autonomous land vehicles so that it's heaving up and down a few metres and see how it goes.

There is a time sequence of events occurring with the boat, it does not go off course at the beginning of the passing of a wave. If the countering helm is applied after the acceleration is first fealt then the action is predictive.

Single handed ocean racers are generally not your common or garden variety of cruising boats.

I can out steer the autopilot when it gets really rough and I can assure you I am not wave watching, I'm generally looking at the compass or feeling the direction of the wind and even with my eyes closed I can tell you "This is a big bugger."

A question for Sean:

The sensor package you use in the Pypilot includes a three axis accelerometer, do you use the output from this and if so how is it used?