The Ocean Race 2023 leg 3: Capetown to Itajaí, Brazil

Herman

Super Anarchist
2,260
1,957
The Netherlands
Interesting, so both ECMWF and GFS give a significant advantage to TM (5 and almost 4 hours respectively), it appears to have paid off to stay east
With currently only 24 nm between the numbers 1 and 2, and loads of weather stuff happening in the coming days I won't take bets yet on TM. Timing of those projected HPs/LPs will be crucial for routing and that will hard to catch by the models, let alone the boats. A bit of rolling the weather dice the coming days.
 

Liquid

NFLTG
5,596
1,295
Over there
Nico seems to be quite the introvert and would happily do it without the fuss of the camera.

There is a photo of them all on the boom and Nico is the only one without his hands in the air.
Rosie is full of bounce,
Will is probably a bit more introverted but seems to come out of his shell a bit - but wants to focus on sailing and winning.
Boris is an old salt (in his 40's).
Nico is the old 'get off my lawn' man who just wants to get shit done
Ant seems to be stoked whenever he gets the drone back which is a bit of fun.
Seems like a good mix of people.

I find it rather humorous. I'm just giving Nico shit for so obviously wanting to be out of the picture. His sailing resume speaks for itself and he's there, I'm here.

But, ya gotta admit Nico does look a little like a hostage here...
m158736_crop110015_800x800_proportional_16799922168820.jpg
 

giegs

Super Anarchist
1,163
667
Do the autopilots have the ability to learn from user input on the fly? Can it recognize that under such and such parameters what type/amount of adjustment is best? Or are the base settings static until it's reprogrammed?
 
Do the autopilots have the ability to learn from user input on the fly? Can it recognize that under such and such parameters what type/amount of adjustment is best? Or are the base settings static until it's reprogrammed?
To be honest they don't often detail much about the abilities of these pilots. That being said, some pilots are customized and trained to a specific boat using detailed models and I believe skipper input is a part of it as well.
As far as I am aware, you can program a pilot to change settings, like "if the wind drops under 15 knot head up 6 degrees" but I am not aware that pilots can themselves take the initiative to change their settings and so on.
That being said, if someone knows otherwise, please do blow my mind.
 
Do the autopilots have the ability to learn from user input on the fly? Can it recognize that under such and such parameters what type/amount of adjustment is best? Or are the base settings static until it's reprogrammed?
I'm an engineer and a pilot as well as a sailor, but not a boat or control systems engineer. However, I think I've learned that at the very least the "gain" or sensitivity of the auto-pilot can be easily adjusted. It can be set to make rudder adjustments more slowly or more quickly, and whether or not to counter-steer ("Meet her") if the boat's course begins to swing past what has been commanded. I would guess - and it's a guess - that these might be compromises between limiting the use of draggy rudder or accepting the drag for more control in higher winds and seas.

I am Artificial Intelligence and Machine Learning-adjacent, and I think they are guaranteed to bleed their way into control systems like the autopilot. However, they need a lot of data to be taught to them and to learn from, and some of that learning will be mistakes. A lot of new tech emerges on very expensive platforms (like racing boats) first before cost and complexity come down for more commercial use. However, I don't think you use the systems first on carbon fiber boats that are doing 30 knots and are 1,500 miles from shore, because some of those mistakes could be catastrophic, so it may be a while before AI/ML is used on IMOCAs doing VG or TOR. But whatever I think, it will probably happen sooner than that.
 
Last edited:

grandsoleil

Member
401
125
Switzerland

erdb

Anarchist
806
619
I'm an engineer and a pilot, but not a boat or control systems engineer. However, I think I've learned that at the very least the "gain" or sensitivity of the auto-pilot can be easily adjusted. It can be set to make rudder adjustments more slowly or more quickly, and whether or not to counter-steer ("Meet her") if the boat's course begins to swing past what has been commanded. I would guess - and it's a guess - that these might be compromises between limiting the use of draggy rudder or accepting the drag for more control in higher winds and seas.

I am Artificial Intellignce and Machine Learning-adjacent, and I think they are guaranteed to bleed their way into control systems like the autopilot. However, they need a lot of data to be taught and learn from, and some of that learning will be mistakes. A lot of new tech emerges on very expensive platforms (like racing boats) first before cost and complexity come down for the "average" user. However, I don't think you use the systems first on carbon fiber boats that are doing 30 knots and are 1,500 miles from shore, because some of those mistakes could be catastrophic, so it may be a while before AI/ML is used on IMOCAs doing VG or TOR. But whatever I think, it will probably happen sooner than that.
I remember reading or listening to an interview I think with Boris or Will about this. I'm pretty sure they use machine learning to tweak the autopilot. They have 100s of sensors all over the boat, might as well use the data! They can also change the behavior of the pilot by selecting limits on certain parameters.
 

climenuts

Anarchist
819
390
PNW
Machine learning is an on-shore process based on off-loaded sensor data. It's too much processing to happen using the power available on the boat. If they're using machine learning they're updating between the legs to be compliant with RRS 41.
 

littlechay

Super Anarchist
1,205
657
Nelson
To be honest they don't often detail much about the abilities of these pilots. That being said, some pilots are customized and trained to a specific boat using detailed models and I believe skipper input is a part of it as well.
As far as I am aware, you can program a pilot to change settings, like "if the wind drops under 15 knot head up 6 degrees" but I am not aware that pilots can themselves take the initiative to change their settings and so on.
That being said, if someone knows otherwise, please do blow my mind.
I don't know about the exact capabilities of these pilots but certainly the boat "learns" to feel the compass (or more accurately the rate and motion sensor; roll, pitch, yaw, heave etc.) and can apply counter rudder etc. before the boat starts to turn; much as a human driver would when they feel the stern lift to a swell and such. The same for heading up and down for changes in wind velocity. That is all fairly standard even on high end consumer APs these days.

As with everything it depends on the quality of the inputs from the RMU, wind gear, plus on these boats the load sensors. I am imagine that the pilot can "feel" the loads and alter course to unload things if it gets out of hand.
 

shebeen

Super Anarchist
I'm an engineer and a pilot, but not a boat or control systems engineer. However, I think I've learned that at the very least the "gain" or sensitivity of the auto-pilot can be easily adjusted. It can be set to make rudder adjustments more slowly or more quickly, and whether or not to counter-steer ("Meet her") if the boat's course begins to swing past what has been commanded. I would guess - and it's a guess - that these might be compromises between limiting the use of draggy rudder or accepting the drag for more control in higher winds and seas.

I am Artificial Intellignce and Machine Learning-adjacent, and I think they are guaranteed to bleed their way into control systems like the autopilot. However, they need a lot of data to be taught and learn from, and some of that learning will be mistakes. A lot of new tech emerges on very expensive platforms (like racing boats) first before cost and complexity come down for the "average" user. However, I don't think you use the systems first on carbon fiber boats that are doing 30 knots and are 1,500 miles from shore, because some of those mistakes could be catastrophic, so it may be a while before AI/ML is used on IMOCAs doing VG or TOR. But whatever I think, it will probably happen sooner than that.
The quote directly from some boat walk through I saw before leg 1 is that the auto pilot uses AI/machine learning to update it's algorithm every five minutes.

Makes perfect sense. Lots of data in, lots of data out. The main goal is speed, more speed. I don't see how it is intangible that a computer can work out how to steer for marginally more speed, despite all the noise.
 

Chasm

Super Anarchist
2,672
523
I'd be quite surprised if noone is doing it already.
Rudimentary machine learning has been a thing in at least one of the self made tiller autopilots for the last 10 years or so.
Simplistic stuff like counting wave patterns. If every 6th wave is different don't treat like the one directly before it.

If you have the computing (& electrical) power there is no reason not to run some form of analysis on board. Computers change a lot between editions. But to do that successfully you need to validate your systems before the race. No idea if teams have invested into that.
 
The quote directly from some boat walk through I saw before leg 1 is that the auto pilot uses AI/machine learning to update it's algorithm every five minutes.

Makes perfect sense. Lots of data in, lots of data out. The main goal is speed, more speed. I don't see how it is intangible that a computer can work out how to steer for marginally more speed, despite all the noise.
From a lot of what I'm reading it's becoming accepted that the auto-pilot steers better than a human can: it never gets distracted and it doesn't need to sleep or eat. I would think the boat senses in six degrees of freedom and the question for the helmsman is probably "How hard are you willing to ride and how much risk are you willing to take?"
 

ET1

Anarchist
542
153
Germany
From a lot of what I'm reading it's becoming accepted that the auto-pilot steers better than a human can: it never gets distracted and it doesn't need to sleep or eat. I would think the boat senses in six degrees of freedom and the question for the helmsman is probably "How hard are you willing to ride and how much risk are you willing to take?"
And it doen´t need to shit in a bucket
 

despacio avenue

Super Anarchist
1,077
352
Alaska
I don't know about the exact capabilities of these pilots but certainly the boat "learns" to feel the compass (or more accurately the rate and motion sensor; roll, pitch, yaw, heave etc.) and can apply counter rudder etc. before the boat starts to turn; much as a human driver would when they feel the stern lift to a swell and such. The same for heading up and down for changes in wind velocity. That is all fairly standard even on high end consumer APs these days.

As with everything it depends on the quality of the inputs from the RMU, wind gear, plus on these boats the load sensors. I am imagine that the pilot can "feel" the loads and alter course to unload things if it gets out of hand.
I recall during the last Ocean Race, you helped out 11th Hour with their dismasting and detour to the Falklands where you put them in touch with your brother and they jury-rigged a mast to get them to Itajai.
 
Top