I think the two seconds delay is a distraction. If your algorithm is measuring the pressure on the foils and sails that will give you enough information to "autopilot" without needing the tracking data. Look at radio control yachts and how well they can sail without anybody on board! The two second delay could be used in a predictive model to guess what is happening to the wind, where other crafts are likely to be etc, but that for me is the role of the crew.
You are allowed to use sensors on the boat equipment but not external data but the boat and sails and the rest of the rigging are effectively indirect sensors of external data. If you "train" these sensors based on external measurements you likely can use them as if they are external. Of course you only get current data and nothing looking ahead but generally autopilot works with current data.
very good point and for me this is the most likely and effective way to use the feedback between controls and loads within the electronic control circuit without having to rely on the delayed environmental data which comes through the media system (and as far as I know is only displayed on supplied screen).
Recent breakthroughs infusion have been brought about by AI recently. Instead of being a few milliseconds behind the development of plasma instability, they are now a few milliseconds ahead of plasma instability. If we had drone sailing boats It could really change the nature of shipping and transportation.q
12:36 combined with accumulated data sets to become predictive. 2 seconds delay is an impediment for systems reliant on only current data. This delay penalty decreases with learning. This is quite interesting
Think of missile targeting. You use past data and algorithms to accurately project where it will be. I am a rocket scientist and the slow dynamics of sailing would be easy to do this.
I love sailing, I love the AC, but truthfully, 32nd AC was the last time this was about sailing, and not about technology and how to outplay one another with hidden tricks.
@@Mickeysternum245 I said a sport we take seriously. Even most of us have come around to the idea that NASCAR is advancing the science of somehow making fast cars boring to watch.
Interesting Mozzy as always. You know more about sailing than I do but there will be ways to infer the environmental conditions from other sensors (sheet loads, mast loads, sail shape, foil load, gyros, accelerometers etc) and then use the environmental sensors on the 2 second delay to actively correlate the data. They did the same with us in Formula E by offsetting the wheel speeds by 2 secs to prevent traction control and effective distance mapping. Wasn't that long until people effectively had traction control by other means and had effectively distance mapped their energy strategies to give a high level of automation. Maybe different in puffy conditions but in a consistent breeze I'm sure it wouldn't be far off autopilot being possible
Yes, I think this is actually more likely that they can use clever feedback between loading on different controls. But all that can be done within the electronic control circuit without needing to rely on the 2 second delayed feed from the sensors that go through the media centre and as far as I am aware is only displayed on one design supplied screens.
I haven't read the rule but wouldn't surprise me if the environmental sensors are on a 2 second delay but even on a 2 second delay this information is useful both for having the boat "learn" live as it can actively compare the two data streams and tune itself and I would have thought on a boat as big as an AC75 a 2 second delay in reality isn't that long and can still be used for a lot of things. For example the boat could work out when it's in a relatively steady state and set the sail shape based off the environmental data live. Would be very different in an AC40 or dingy which you're tuning big amounts constantly but when an AC75 is up to speed and three times wind speed it's actually generating a lot of its own loads.
Hi, about the two seconds delay of the control system output: Data logged by the system can produce forecasting based on the history. So even if the system uses , at a given instant, last sensor data of two seconds before, by using a model projection can act in actual time or even in anticipation of the change… given the model capable of predicting two seconds ahead, which is possible in waves and wind patterns.
Anticipate one way or another to spread out the task. Determine a proxy indicator/event that is 2 seconds earlier in the stack & use it. On a 75 the scale of the response may be significantly less time critical than on smaller vessels.
All the teams are using automated dynamic controllers to fly the things. Have you ever seen small boats flying on hydrofoils? They bob and jerk all over the place especially in transition.
Add a feedforward loop to make predictions based on tested or derived data, and I think the 2 second problem goes away. It can be done, but let's face it, that's not sailin' :)
@@mrhyde3029 but it may be winning , look what happened when ETNZ wer 8 1 up of the Spitbull and Oracle cracked the code on foiling , instant change of the dominant boat , it was so much smoother overnight
Check this guy out. He understands the game and he asks questions of the narrative being presented, unlike any of today’s MSM wafflers. Good analysis and honest. Bravo 🎉
Great presentation, thanks. For me the 2 second rule does not preclude a constant automatic update of the mode of the yacht to keep it in the envelope that makes it controllable. A modern fighter jet is dynamically unstable and hence highly manoeuvrable. It is uncontrollable until a computer automatically inputs control deflections to stabilise the platform that enables the pilot to fly it. Foils on the edge of stability for a human input could be kept in a window of human controllability by constantly updating the base setting of the yacht, making it controllable and fast. Mercedes should be all over this as a race car is a similar beast.
Holy hell, I’m not really a sailor, I’ve done the Chicago to Mack a dozen times, I grew up sailing but haven’t been on a boat in years, but these new AC boats are bonkers! I got a chance to go out in one of them back in the early 2000’s and man, have things changed! That’s really amazing.
There might be some aspects of trim where a couple of seconds latency would not be catastrophic? I'm thinking of the equivalents of cunningham, outhaul, rake, ride height etc on a dinghy. Can the boat "auto-tune" itself to the prevailing conditions, while the crew make fine-tune adjustments around that setup? Presumably there will already be some degree of "fast" automation to help deal with (e.g.) gust response without relying on data from the ILS (i.e. naively perhaps the boat senses the sheets being eased together and the steering heading up into a gust and automatically flattens the rig and adds a bit of twist for a second or two?). Fusing the delayed data stream from the ILS with some inference based on low-latency data from the ECC would not be trivial, but it does sound feasible for a team with plenty of smart people. I guess you wouldn't need to aim for full autopilot, just taking a bit of cognitive load off the sailors and making the boat a bit friendlier to sail "on the edge" would still be a valuable thing?
With the power these learning algorithms and hardware have shown in simulating fluid dynamics, processing the past 60 seconds based on the inputs/sensors and what all of the hours of training predicts, a 2 second delay is nothing.
As a software engineer, I don't think it's that much about the two second delay you mentioned. That delay is built in specifically to prevent that data to be used for automation. What I think is quite possible is that the teams could add scripted automations that include arbitrary additional delays. E.g. you start the tacking procedure, which initiates the lowering of the windward foil. As soon as the foil is down, it prompts the helmsman to turn the boat, after which he confirms by pressing the button again, initiating the new windward foil to be raised, the mast to be turned and whatever else needs to be done in exact order. Similar scripts could be implememted for gybing and other manouvers. The mentioned delays would be in between some of these actions. Maybe the foil only needs to be raised 0.3 seconds after the mast was rotated, which is quite simple to automate.
This could also include something like continuously change foil angles during a tack where the new leewardfoil is set to maximum lift initially, but then changed to the original position of the old leeward foil gradually over time. Not sure whether that particular example would be useful, but just as an example what could be possible.
Automation background and think this is exactly what will happen but even deeper. I’d say they will have presets for all their manoeuvres up through the wind range,tacking, gybing, bearaways, round ups, Jks, anything that has to happen in a sequence will be automated. Foil/sail adjustments could still be automated after each manoeuvre until they know they are up to target speed before the sailors take control again. The only time I see the sailors in control is when they are going straight looking for speed and reacting instantly on feel to the boat and wind.
@tonyward2101 No, I'd think you'd want to choose that by feel and understanding. They do have live data on board like on any other boat, I'd assume. And it makes a lot of sense for the sailors to know exactly when they want to initiate a manouver.
@michaellynch9284 Agreed. The interesting thing will be to see how far you can get with this kind of automation and where the sailors have to take over. E.g. I assume that the rudder and potentially the sheets cannot be controlled automatically. And even if they could, I assume a human expert would still be better at trimming the sails during a manouver than any preset script without any information on the conditions. The second the machine gets all the measurement data and a few cameras we're in a completely different situation.
What are the limits on using a database to predict two seconds into the future based on the previous 3:00 to 0:02’? Also, you can think of some of the automation like and engine map in F1 or motor racing, like how the throttle pedal responds given the fact that you’re speeding up after a tack or are likely preparing for one.
I liked the introduction to foiling when it was introduced in the 34th... but all of the automation has just turned it into a battle of programmers and not really sailing anymore.... the only sensors they should have are the wind data, and location on course, and let the sailors figure out the rest on their own.
The 2 second delay can be overcome by an "autopilot" if it has fuzzy logic that can learn to use the time history, including the delay, to extrapolate forward into real time or the future. This will require the aquisition of vast amounts of data for the logic to use to learn from, but is very possible with a sufficient degree of accuracy to provide a base level of boat control, to which the crew can monitor and refine the final response in real time.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen (all one design supplied equipment). The crew reads that screen and can then input signals (press buttons, turn knobs) at will.Any fancy logic has to happen downstream of the crew input commands.
That could be problematic. The data can be re-acquired via a camera, and the availalble measurement data from the vessel systems can be used as additional input to guide the extrapolation, but it is certainly less than ideal.
They've also written the rules so could have 'looked at it' years ago. I've no doubt they start with innovations already in their locker and write the rules to allow them without making it obvious to other teams.
@@TVRV8 I think this is the conspirator element of the story. ETNZ and INEOS wrote the rules. Did they use that power to to conspire to include an area only they had the resources to exploit? That seems to be the inference behind what Terry Hutchinson is saying
@@MozzySails The simple fact he's suggesting at it makes you KNOW he would have done it IF he could have so .... All is well that ends well. Instead of "complaining" and suggesting they should be working on making their boats go faster. Leave politics ashore and to the politicians.
Welcome to the world of America's Cup conspiracy theories 😁 Not sure if it was confirmed, bit was certainly rumoured that Blair Tuke was tapping a dot on a screen whilst aboard his bike. Tapping the dot would break the 'computor autopilot' loop adding a human step where taping the dot would tell the computer to move the foil rake to the desired angle of attack.
Could you not train a model to predict what the wind speed will be in 2 seconds time? I'm not sure whether that could be at all accurate. Or perhaps you could have the sailors providing real-time input to guide the predictions. How long do these procedures take? Perhaps the model predictions are good enough to start the procedure and the sailors finish it off?
Exactly my thought. A derivative model using a continuous stream of data should be able to do that. Algorithms based mathematical models such as the Black-Scholes used in options pricing should be able to give you at least some advantage over flight specialist intuition.
Yes you can. I am coming from a Wave energy converter tech background. In this field, to extract the most energy from the waves, you need to control the force apllied by a Power take off (link to generator in most cases) to the prime mover, to control the system dynamic. Control theory gets better if you can provide your system with future wave loads, i.e. if you can predict what will be the loads in 5, 10, 15s. It is something you cannot do analytically, but with the help of machine learning, you can definitely predict the wave loads at a 10 to 15s horizon using all your device state parameter. So 2s delay will still give you some good level of forcasting. I am sure the application is different, but in an environment with waves with mean periods >5s, gust also lasting in seconds, it might be doable to train a system giving you more than 2s forecasting, therefore it might help anticipate some controls on the yacht, and not be reactive only
In vehicle ECUs the industry uses look up tables which is logged and stored from real time data (normally used as a fall back). This can be pre gathered and rendered down. a 2 second delay will allow you to get into a ball park, with this you can use well known deterministic algyrythms to best guess the next set of numbers. This logic in machine control systems was developed as standard practice in the 90s. These (2010+) days everything in controller world is more sensor driven with target and actual values. EG you have a target and your closed loop system is reading the actual, it then has to work out how to get the system you're automating to the target value. If sensors arent allowed in realtime .. the near realtime could still be used with prediction to calculate the best guess value and replace that missing data with its best guess (guided by the rendered lookup table from previous data). Its what I would do anyways if I wanted to build a system like this within the ruleset.
Thinking out loud here, but if you had a sophisticated enough state estimator (math model) of the boat and wind conditions based on IMU or other legal sensor data, you could use it to output a 2s-delayed AWS value that the crew then compares to the telemetry data available. Then you just get them to adjust the up/down knob based on how accurate the estimator is travelling compared with the telemetry to give the model enough feedback for the Kalman filter jigger inside the box to do its thing.
Hey I don't think 2 seconds are a lot when considering the mass and inertia of an AC75. And additionally, as many others pointed out, 2s delayed information is still amazingly useful for a computer that knows perfectly the behaviour of the boat and has a complete history of the status. In my opinion once you allow electronics you are guaranteeing the ability to design an autopilot. Maybe that is very advantageous to the defender, making the learning, designing and cost curves much steeper for the challengers
You can probably do this with a trained AI model. There is a natural rhythm of the boat, even across the two second delay, that can be understood and leveraged by an AI model enough to improve the handling of the boat, particularly if you capture enough training data from human control during the testing phase. This would be an interesting discussion to open, and probably worth a video on the adoption of AI technology in sailing.
nope, he's talking about automatic operations that are carried out in sequence, with a pre-defined delay. This delay(s) could be adjusted on parameters that are not boat external (actuators pressure and position, ecc...)
I’d think of it less as a conspiracy theory and more like a fact. Surely they can script maneuvers and setup automations that would be useful despite a 2 second delays. Not to full on autopilot levels hence “quasi”
If you could identify a set of specific operations and conditions, like tacking in a reducing wind, or dipping another boat, or pinching to reach the mark, for example, you could use machine learning to collect data and associate it with these events. Then you could potentially optimize either the automated interaction between internal systems (foils to mast rotation, for example), or even the sensitivity or linearity of the manual controls, to improve effectiveness for future examples of similar events. Does the rule prescribe limits to on-board computing power and alteration of behavior during a race? I'm not sure I'd call this an autopilot, but it is a refinement of the interaction coefficients between different controls on the boat, which could perhaps increase the performance of the boat over the course of a race or a series. This could be especially useful in outlier conditions, but it comes with the risk that the boat's behavior might not match the expectations and experience of the sailors, and therefore be counterproductive. Used with care I could imagine this being useful.
I think it would be possible to use the data with the delay in some situations, say trim optimisation or small adjustments on rudder position to head the boat up into a lift. The reason that I think it would be possible is that they would not really look at the absolute data, what I think would be more useful is looking at the trend in data that humans don’t really have the ability to use. If you think about when you are helming, you have a variety of inputs, you can see the colour of the approaching water and as you get to the edge of a wind band you can make significant adjustments to trim or rudder input to take advantage of the change in wind speed and direction, but it may be that if you look at the data, the transition isn’t as sudden as you think and there are some clues that you can see from the data if you have a richer data set. To some extent as a helm or crew of any performance dinghy there is a lot going on and I expect that everyone becomes overwhelmed with information and inputs, but if you are an electronic device that doesn’t have to listen to the other person in the boat yelling at you about why you called a lift when it wasn’t or suddenly having a face full of water as you crash through a wave, then you can have a much more objective review of the data. Anyway, I’m not in automation design for americas cup and I’ve not actually looked at the data the external facing systems produce that’s delayed in that way.
exciting. I’ll also spray some cologne on the fan. No one prohibited visual communication with the sensor. It is enough to place a millimeter scale (sorry, inch) next to the direction sensor. Next to the speed sensor, a tuning fork. and view them via gopro. the speed of light is the maximum speed possible in this space.
Sensors showed the controller where to move the ride height, dot on screen moved and controller mover joy stick to follow it. Clever, at least it was not cheating like Oracle did in San Fran with Herbie.
Having ANY electronics control doesn't do it for me. I would also like the cup to be ran regardless of wave height. If a team and boat is better then it shouldn't need precise conditions in order to run/compete.
So often with sophisticated techniques it is the use of (probably hidden) layers that are outside the comprehension of those without due expertise. We as outsiders may be stuck in mindset of a single layer that does it all, rather than the interaction of multiple layers each doing a smaller thing with predictability with all scenarios mapped so as to avoid the unforeseen. Reduce the cognitive load on those pushing the buttons and seek some small advantage. What weaknesses of the boats might they be seeking to target because those limitations may more fundamentally create the character of these vessels.
If all that is required is for the mast to rotate when, or just after, the foil is dropped or raised then no feedback is required at all. It's just one switch doing 2 functions which is not difficult to achieve - with or without a computer. It's barely even automation. If I flick switch A then the port foil drops. If I flick switch B then the mast rotates. Or if I flick switch C the foil drops and the mast rotates. Where is this claimed non-permitted feedback?
There is no claim of evidence of non-permitted feedback. I just showed the mast as evidence of what was allowed and think it's a pretty basic relationship and now really that much more sophisticated than linked controls on dinghies (spinnaker goes up and pole goes out on same line). The claim that something else can be done is purely speculation from Terry Hutchinson. I am just outline the technical context for it and am sceptical myself it is possible to achieve.
Great vid as always. I think it is a good question to ask for every control/adjustments on the boat, can a human see something and react in significantly less than 2s rather than the computer seeing and reacting with a 2 sec handicap. I suspect there are many controls that a human can’t react significantly faster and so thereby gives both a cognitive advantage to the humans by load reduction and/or more precise changes where automation can do better.
Excellent thank you, great to know about mast turning automation etc, I'm as excited as you are to see the new boats, flights and accom are booked 12 - 20 Oct can't wait, go ETNZ
No surprise that team NZ set up the rules in a way they can game them. The automation will spell the death of the event as a sporting spectacle that's supposed to be all about sailing skill.
The boats are big with a lot of mass. 2 seconds can be productively utilised I'm sure. There will be plenty of of lead/lag power relationships on board. Being able to dynamically manually adjust those relationships(AWS +,-?) at different speeds would be key though, i.e. powering up the mast angle 2 seconds after the middle of a tack & balanced foil cantilever angles. As a kiwi myself, I have to say that it is in the nature of the America's cup rules and history to use boat design rules tactically against the opposition. Were it me designing the control algorithms by the permitted means, whether fluidics, silicon logic or other, the goal is to put the hydraulic operation sequences into a natural harmony. We have history in other fields, for example Fords when their designers were still learning how to write software, of loosing inertia for flywheel mimum RPM retention under light injection and intentionally dropped cylinder's combustion sequences(to maintain some stoichiomeric mix in active cylinders at partial loadings). They fixed that with crank angle sensors and presumably a SW update. I knew because I'd invented my own engine with the same design goal practical implementation concept. To put those sorts of factors under manual control is a bad idea, you have high incident repetition with tight timing, it's not well suited to human control(the Ford would eratically stall on take-off). So I welcome this sort of rule change. It's forward thinking and analysis like you are doing is potentially very rewarding. I think the means of implementation of the logic is irrelevent, that horse has long bolted the stables. Those who criticise these high speed foiling works of advanced engineering teams yearn for simplicity and low IQ requirements gear but that's not what this cup racing is about.
i've always said this inconsistency in contemporary AC racing. There is no checks on the software load. It could be full of cheats. One fix: Have a configured software load that goes into the boat right before the race. Teams must submit their source code in advance of the race.
If you have systems that can react fast enough, could you theoretically use sensors on the jib's control systems as a pseudo wind vane to automate certain aspects of mainsail trim? Since these boats are basically always sailing upwind, the jib will always experience changes in wind angle and speed slightly before the main does. I would imagine that these "3D sheeting systems" that some of the teams are using on their jibs would allow for more accurate data collection from the jib itself due to the forces being distributed across multiple lines that could each have their own load sensors on them.
I think that he’s on a fishing expedition. There’s a vast difference between automation of linked systems functionality to reduce crew workload (improving safety by reducing critical systems operating c^ck-ups) and an autopilot that will get you from point A to point B without human input. The 2 second delay would effectively render any such implementation less efficient at the task than the human crew, doubtless why that stipulation was made. Having spent time as an engineer in motorsports during the transition period from carburettors to fuel injection, I can hear echoes of a number of objections that did the rounds at that time. They were the exclusive preserve of older and more dogmatic individuals who were, quite frankly, out of their depth in terms of how the technology functioned and what it’s capabilities were. After that frustrating experience I’m just glad that I’m not even remotely involved in attempting to police these regulations, or indeed to defend the innocent against one unfounded accusation after another.
Not that familiar with the Americas Cup control systems or rules but when we did your foiling electric boat there is definitely a way to use the 2 second delay information to tune the system. We would run test and simulation systems using hardware in the loop and then use the data coming out to then modify the gains and derivatives on the controller to improve the flight response to conditions and optimize. So if this all they are doing on the water then your comment @mozzysails would make sense that they had buttons with + and - for various variables as the data comes through they can then make the changes based on the new info coming in even if its 2 seconds delayed. One of the best tools we used was to have the ability to plot data and then use that to see if the changes were moving in the right direction or not, so by overlying the right data you can see if you are making efficient decisions. So in essence its not automation but more semi manual manipulation of the system to give better performance. The 2 second delay could also be giving feedback on how much your settings were out and from this you can then know how much your "feel" for the boat is going, so for example you trim the foils and them 2 seconds later you know what the performance output was. When you get used to this method it almost becomes "normal" as opposed to being delayed.
One thing to consider is that TNZ utilised AI to drive their simulator in the last cup. It wouldn't take much for the internal systems to now have an AI component in charge of the automation. You could train the AI in the simulator and then upload the latest trained AI to the boat. It would be like having an additional artificial crew member.
Can you make a video on the "follow the dot" program by New Zealand in 2017? How did it work? I searched for info, but only find short mentions of it online. What is known?
2 second delay is perfect for working out the sea state. If you use cap sensors on your rudder you can get mm accuracy at what ever bus speed you can read. Much better than ultrasonic which is like 10x per second cm accuracy.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen. The crew reads that screen and can then input signals (press buttons, turn knobs) at will. So the control circuit which can link controls can't be running on actual data with fancy algorithms.
@@MozzySails yes but can they process that data first? if they can, they could use that to establish the sea state, wave heights, periods and also if your on a lift or a knock if there's mcu data. The thing with cap sensors you can embed them in the hull. So if you can use that data even on a 2 second you can establish targets that can then be used with the allowed live data from load cells and do a chase the dots. (I made a flight controller and was switching to use cap sensors when covid hit and I gave up as a lost cause)
@@andrewferguson8735 I don't believe so. If you pause the video at 7:19 you will see in the top right of the diagram that everything in grey is a supplied system that is the same for all teams. That allows data to appear in the Crew Information System (CIS) and display on the crew indication devices. The only thing I can think of is that the crew induction device could process the data and display recommended settings to the sailors. You will see there is no link from CIS to the Electronic Control Systems during racing, so any commands that go in to the ECC must come either from the crews input devices (buttons etc) or from sensors within the ECC.
I think if an AC team really wants to put doubt and a fright into the other teams, they should hire the former CEO and the engineers of VW who masqued the diesel software emissions controls to dupe the regulators. Could an AC team rig their electronics to fool the Official AC boat measures? Maybe. 😃
Agreed , bring back wooden hulls cotton sails, no life lines open cockpits and a proper skipper wearing a peaked cap and a rolled neck sweater, smoking a pipe of ready rubbed St Bruno tobacco. Every boat should be free of advertising and named after a sea creature or exotic wind events. Bare breasted carvings on the prow hand painted by students of the Salvador Dali school of art. Oh last thing, no black lesbian trans volunteers to be allowed a place on the crew without a note from their mum. 😅
Exciting stuff, but technology has really levelled the playing field and to be honest, the racing gets a little tedious. Think I preferred it when it was more reliant on manual skills
Good question. Am from newzealand do think newzealand has had this innovation on point from the get go am thinking they would have worked it back to stay legal will at least hope they did .they will have lots of options that will be available questionable to say the least. Guess they're waiting to see if needed. As will have to explain if do .
2 second delay is ok if they are also allowed to run predictive algorithms. It's like weather forecasting in microcosm, no? You just compute the differentials, and run the dynamical system (or these days incorporate a neural net for some of the number crunching). Means they have to gather a lot of data at the venue _beforehand_ to train the ANN in advance. You'd have to prove me wrong to discount this. A sailor "looking well ahead" can probably be beaten by a decent ANN if the training data match the current weather well enough. Might only be a minuscule advantage, but... this is the AC.
So here's the thing. The data displays are supplied equipment. So as I see it, it's a one design link from measurement devices for environmental factors (wind, speed, ride height etc.) with a 2 second delay. The crew then reads from that screen and presses buttons. So in terms of fancy linking of controls, that all has to be done in the electronic control circuit downstream of the crew inputs. I could see if you could get the environmental data in to the control circuit, then have some fancy algorithms which forecast future best responses, then you may get something that works even with the delay. But I don't think this is possible with the 'human link'.
Great update. Automatic sail and rig trims seems sensible, but any update that brings "less fun" will surely end up being self defeating. Porsche 911 in manual gearbox spec shows what the market thinks here. Also, Is the term "gybe" now superfluous and unnecessary? Given awa, why are we not just tacking downwind?
Very complicated boats, and it seems that this means there are going to be grey areas. For instance they need to have control systems for the ride height that react faster and better than any human. Is that considered a kind of autopilot or not? We know that there is far too much data for the helmsman and strategist to respond to, so automation has to come in otherwise these boats are unsailable. Another complication is that once the computers have been set up, it is virtually impossible to read the code back to figure out what exactly it does. Will the organisers have to scrutinise the code on every boat to ensure that there are no transgressions into sailing autopilot technologies. I'm glad I'm not a race organiser. I'm sure as a AC skipper, what they would love is a boat that largely sails itself, barring a few nudges on the wheel and a couple of button presses per tack, so that they can focus out of the boat on what's happening tactically on the race track. Is this we want out of a Yacht Race? When I started sailing, all I had to coordinate were two ropes, centreboard height, body position and the telltale in conjunction with reading the wind, waves and current. It was very visceral and physical, and mostly quite slow. I do enjoy watching these Formula One boats racing, I like innovation but I don't want to see cheating.
I thought that anything above the waterline had to be powered by human force, while anything below could be battery powered. If that is the case, how can the mast rotation be automatically synchronized with the battery powered movement of the foil arms?
you're correct about the power sources, but signals to both go through the same electronic control system so can be linked there. There is a degree of stored power in the human system vi hydraulic accumulators, but essentially the valve opens to turn the mast and as long as the cyclors have built up enough pressure for it to move, then it will move.
Americas Cup is no longer for sailors, but for engineers and programmers... The spirit has been corrupted, engulfed by the need of win for the sponsors. Don't get me wrong, development is always a must, but I would not allow a single piece of electronics onboard. A handheld radio and that's it.
yes, planning on getting down to Weymouth to have a go. Looks cool. I like the concept of one design which freeze in time a peak from a development class
And here I am sailing a boat designed in the 70s built in the 80s that spends most of its time being steered by a mechanical system that was pioneered in the 30s (Moody 37, with wind vane steering) She ain’t fast, and she doesn’t keep to the rhumb line but it usually averages out pretty good, and she’ll get you there… eventually…
You are assuming that the data they're collecting is raw and from a sensor feed or array of sensors. Most likely it would be from the inputs made by the crew, in the form of certain conditions, like Wind speed, angle of attack, current sail and daggerboard conditions, etc. In that case it would not have to be a 2 second delay. Not necessarily an Autopilot, but a “Sailing Assist”
The 2 second delay at the speeds these “boats” are going changes the evaluation. There should be NOT any digital connection or “automation” or adjusting or movement of “systems” for sailing
What if there is a 2s delay from sensor to execution system, but there is a machine learning predictor logic, trying to “learn the future” and having an instant (2 second) feedback loop. And the execution system makes decisions on the basis of the prediction.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen (all one design supplied equipment). The crew reads that screen and can then input signals (press buttons, turn knobs) at will. Any fancy logic has to happen downstream of the crew input commands.
Is it an autopilot if it shows you the optimum setup and a human follows that setup? The two second delay may be the delay from switching from one mode to another. For instance foil trim and then sail trim/mast position. In the video showing the foil dropping and the mast turning, there's a two second delay between the actions. Is that the loophole that Team New Zealand are exploiting?
High quality analysis as always. Thank you. One thought - is there any way AI could learn to predict the direction of change from a two second delayed feed. If I saw the same thing happen ten thousand times followed by a predictable change, couldn't I effectively eliminate the delay with predictive logic?
I think you could run AI on it during training and in the simulator. Some sort of machine learning really that can then produce algorithms to be used in the racing to coordinate control... but I still think the 2 second delay would make it really hard
I wonder if the idea is that computers could be used to "simulate" what would be ahead on the race course, essentially predict the present and use that for a quasi autopilot
There isn't a lot on it, despite what felt like lots of discussion at the time. You can read this article on it here: www.sailingscuttlebutt.com/2017/06/19/americas-cup-stick-a-fork-in-it/
I think it's a lost cause disallowing autopilots but allowing "blind" autopilots. What you will get is crew control becoming the environmental input for a neural net. There will be no end of accusations and counter-accusations when it comes to that input into the autopilot. One would like to turn back the clock and disallow feedback control, but that ship has sailed, pardon the pun.
3:20 "It's pretty clear irrefutable evidence" It's not though, it's not even bad evidence, it's just a video that shows the boat doing two things. it's _you_ who insists that the mast cannot rotate by any other means than automation 'because why would anybody rotate the mast while not moving?', except you just explained that lowering a foil is _always_ followed by rotating the mast so why is it now suddenly strange that they did that here? It's what they train for... What you see in the video literally is the routine they go through when lowering a foil, always, on every boat. It would be evidence if the guy in the cockpit tells us all he did was push a button to lower the foil and then the mast rotated, but that's not in the video. And even then we can get into all kinds of nooks and crannies about things like having a button that has two contacts; one to lower a foil and a second to initiate a mast-rotation after a five second delay, which would give the same result but is not automation. Not saying they do this, obviously, but my argument is that the video is does not show what activated the mast rotation.
yeah, going from memory of what I saw last year! Realised it in the edit, probably should have put a note in. I think the principle still holds. They also have a UW/DW button which could be to switch configuration for each leg. Then the green and red button pressed for each tack to rotate mast and raise board.
the amount of sailing and wind data these guys have and the data analysis capabilities available to them makes me wonder if theyve been able to find very fine and small patterns in the wind around barcelona, and then use these patterns to be able to predict 2 seconds into the future? maybe a coarse autopilot is available from this automation rule and processing power? something like steer assist in a car racing game like forsa or similar. its a wild thought.
but what if you have the human subjectively evaluating the wind speed, looking at gusts etc, not looking at delayed sensor data but looking at waves etc. not accurate you'd say but if you spend time training a number of professionals and selecting the best ones at it I reckon it might be quite accurate?
I don't even understand why we're still able to follow this competition!... What happened to the sail changes, the gybes under spinnaker, the luff starts and everything else that made this competition so beautiful? Today, it's just six guys pumped up on protein, grinding away non-stop and 'water snatchers' speeding along at 40 knots on a lake.... We're bored!!!
With modern AI it might be possible to train a neural net to forecast the values based on the current ones. To try to minimise the 2 second delay in the data.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen (all one design supplied equipment). The crew reads that screen and can then input signals (press buttons, turn knobs) at will. Any fancy logic has to happen downstream of the crew input commands.
Possibly the autopilot won't so much be an autopilot but a man in the loop synthetic control system. In the airbus for instance the pilot makes control inputs but the aircraft then auto trims to hold the aircraft in that new attitude. Perhaps they will be able to reduce the load on the trimmers. Always running 2 seconds behind does seem a problem. Possibly it would be more of a back up for when a sailor hasn't had time to do something yet.
As a person just watching this as a sailing fan, this is very dissapointing to learn. They should ban electronics from the AC, and make it about sailing and sportmanship again.
I think it’s totally doable. If I were working there, I would implement Kalman filters to predict all the environment information at present given them at 2 sec in past. Straight line speed optimization is totally doable here. Esp. With a sailboat and a known 2sec delay. It’s not rocket science, it’s known robotic science.
I suggest that in the Formula One world there is an underground river about those themes and the battle for win is decided day after day more by electronic and sensors and AI than an engine or driver ability
Data logs fine tune the boat, looking at data and adjusting the boats tune makes them faster in certain conditions, just like car racing. ecu tuning, different tires and pressure for different tracks/conditions. car doesnt know how fast the winds blowing or which way . the driver has to keep the car straight and adjust accordingly, speed, steering wheel etc.
I think the two seconds delay is a distraction. If your algorithm is measuring the pressure on the foils and sails that will give you enough information to "autopilot" without needing the tracking data. Look at radio control yachts and how well they can sail without anybody on board!
The two second delay could be used in a predictive model to guess what is happening to the wind, where other crafts are likely to be etc, but that for me is the role of the crew.
You are allowed to use sensors on the boat equipment but not external data but the boat and sails and the rest of the rigging are effectively indirect sensors of external data. If you "train" these sensors based on external measurements you likely can use them as if they are external. Of course you only get current data and nothing looking ahead but generally autopilot works with current data.
very good point and for me this is the most likely and effective way to use the feedback between controls and loads within the electronic control circuit without having to rely on the delayed environmental data which comes through the media system (and as far as I know is only displayed on supplied screen).
@@luciustarquiniuspriscus1408 2 seconds
Recent breakthroughs infusion have been brought about by AI recently.
Instead of being a few milliseconds behind the development of plasma instability, they are now a few milliseconds ahead of plasma instability.
If we had drone sailing boats It could really change the nature of shipping and transportation.q
We already have full autopilot on commercial vessels and pleasure motor yachts.
12:36 combined with accumulated data sets to become predictive. 2 seconds delay is an impediment for systems reliant on only current data. This delay penalty decreases with learning. This is quite interesting
Think of missile targeting. You use past data and algorithms to accurately project where it will be. I am a rocket scientist and the slow dynamics of sailing would be easy to do this.
this is why the americas cup is so epic and will never be replaced by the sail gp. development, development, development.
Sail GP also doing alot of development, but they are more focused on the sailors sailing the boat and not a alot of algorithms
@@shiraz1736 accept and embrace change!! Old school sailing still excist
@@shiraz1736 that comment also tells me you dont understand either sailGP or americas cup
@@shiraz1736 and you stuck in the past!! Bet your sailing life is boring as fuck
I love sailing, I love the AC, but truthfully, 32nd AC was the last time this was about sailing, and not about technology and how to outplay one another with hidden tricks.
Looking forward to hearing from you more frequently in the next few months
My favorite sailing show. Well done, Tom.
Cheers
Many thanks!
American Magic don't have the capabilities of little old NZ? Crazy...massive compliment to the ingenuity of the kiwi team
This has been the case for years. Remember KZ7? The 'murricans are still trying to catch up.😂
Fun fact: one of the kiwi bankers who initially financially backed the team is literally named "David Richwhite" lmao.
@@mattyb7736 Beat us in any sport we actually take seriously as a country.
@@ev6558 Like when Shane Van Gisbergen won a NASCAR race on debut?
@@Mickeysternum245 I said a sport we take seriously. Even most of us have come around to the idea that NASCAR is advancing the science of somehow making fast cars boring to watch.
Interesting Mozzy as always. You know more about sailing than I do but there will be ways to infer the environmental conditions from other sensors (sheet loads, mast loads, sail shape, foil load, gyros, accelerometers etc) and then use the environmental sensors on the 2 second delay to actively correlate the data. They did the same with us in Formula E by offsetting the wheel speeds by 2 secs to prevent traction control and effective distance mapping. Wasn't that long until people effectively had traction control by other means and had effectively distance mapped their energy strategies to give a high level of automation. Maybe different in puffy conditions but in a consistent breeze I'm sure it wouldn't be far off autopilot being possible
Yes, I think this is actually more likely that they can use clever feedback between loading on different controls. But all that can be done within the electronic control circuit without needing to rely on the 2 second delayed feed from the sensors that go through the media centre and as far as I am aware is only displayed on one design supplied screens.
so it surprised me that Terry mentioned this 2 s delay.
I haven't read the rule but wouldn't surprise me if the environmental sensors are on a 2 second delay but even on a 2 second delay this information is useful both for having the boat "learn" live as it can actively compare the two data streams and tune itself and I would have thought on a boat as big as an AC75 a 2 second delay in reality isn't that long and can still be used for a lot of things. For example the boat could work out when it's in a relatively steady state and set the sail shape based off the environmental data live. Would be very different in an AC40 or dingy which you're tuning big amounts constantly but when an AC75 is up to speed and three times wind speed it's actually generating a lot of its own loads.
Hi, about the two seconds delay of the control system output: Data logged by the system can produce forecasting based on the history. So even if the system uses , at a given instant, last sensor data of two seconds before, by using a model projection can act in actual time or even in anticipation of the change… given the model capable of predicting two seconds ahead, which is possible in waves and wind patterns.
Anticipate one way or another to spread out the task. Determine a proxy indicator/event that is 2 seconds earlier in the stack & use it. On a 75 the scale of the response may be significantly less time critical than on smaller vessels.
All the teams are using automated dynamic controllers to fly the things. Have you ever seen small boats flying on hydrofoils? They bob and jerk all over the place especially in transition.
Yes a new video to consume instead of sanding antifouling
You can use a PID loop to post check your internal data. It would allow for error correction of the current input.
Add a feedforward loop to make predictions based on tested or derived data, and I think the 2 second problem goes away. It can be done, but let's face it, that's not sailin' :)
@@mrhyde3029 but it may be winning , look what happened when ETNZ wer 8 1 up of the Spitbull and Oracle cracked the code on foiling , instant change of the dominant boat , it was so much smoother overnight
This is basically a kalman filter, and is a well known technique in control systems (like autopilots)
Check this guy out. He understands the game and he asks questions of the narrative being presented, unlike any of today’s MSM wafflers. Good analysis and honest. Bravo 🎉
Great presentation, thanks. For me the 2 second rule does not preclude a constant automatic update of the mode of the yacht to keep it in the envelope that makes it controllable. A modern fighter jet is dynamically unstable and hence highly manoeuvrable. It is uncontrollable until a computer automatically inputs control deflections to stabilise the platform that enables the pilot to fly it. Foils on the edge of stability for a human input could be kept in a window of human controllability by constantly updating the base setting of the yacht, making it controllable and fast. Mercedes should be all over this as a race car is a similar beast.
Holy hell, I’m not really a sailor, I’ve done the Chicago to Mack a dozen times, I grew up sailing but haven’t been on a boat in years, but these new AC boats are bonkers!
I got a chance to go out in one of them back in the early 2000’s and man, have things changed! That’s really amazing.
There might be some aspects of trim where a couple of seconds latency would not be catastrophic? I'm thinking of the equivalents of cunningham, outhaul, rake, ride height etc on a dinghy. Can the boat "auto-tune" itself to the prevailing conditions, while the crew make fine-tune adjustments around that setup? Presumably there will already be some degree of "fast" automation to help deal with (e.g.) gust response without relying on data from the ILS (i.e. naively perhaps the boat senses the sheets being eased together and the steering heading up into a gust and automatically flattens the rig and adds a bit of twist for a second or two?).
Fusing the delayed data stream from the ILS with some inference based on low-latency data from the ECC would not be trivial, but it does sound feasible for a team with plenty of smart people. I guess you wouldn't need to aim for full autopilot, just taking a bit of cognitive load off the sailors and making the boat a bit friendlier to sail "on the edge" would still be a valuable thing?
You don’t need the delay, you can project with past data and filter. Still better than human.
With the power these learning algorithms and hardware have shown in simulating fluid dynamics, processing the past 60 seconds based on the inputs/sensors and what all of the hours of training predicts, a 2 second delay is nothing.
As a software engineer, I don't think it's that much about the two second delay you mentioned. That delay is built in specifically to prevent that data to be used for automation.
What I think is quite possible is that the teams could add scripted automations that include arbitrary additional delays. E.g. you start the tacking procedure, which initiates the lowering of the windward foil. As soon as the foil is down, it prompts the helmsman to turn the boat, after which he confirms by pressing the button again, initiating the new windward foil to be raised, the mast to be turned and whatever else needs to be done in exact order.
Similar scripts could be implememted for gybing and other manouvers.
The mentioned delays would be in between some of these actions. Maybe the foil only needs to be raised 0.3 seconds after the mast was rotated, which is quite simple to automate.
This could also include something like continuously change foil angles during a tack where the new leewardfoil is set to maximum lift initially, but then changed to the original position of the old leeward foil gradually over time. Not sure whether that particular example would be useful, but just as an example what could be possible.
Automation background and think this is exactly what will happen but even deeper. I’d say they will have presets for all their manoeuvres up through the wind range,tacking, gybing, bearaways, round ups, Jks, anything that has to happen in a sequence will be automated. Foil/sail adjustments could still be automated after each manoeuvre until they know they are up to target speed before the sailors take control again. The only time I see the sailors in control is when they are going straight looking for speed and reacting instantly on feel to the boat and wind.
… and they can use 2 second old information to chose which script to run…
@tonyward2101 No, I'd think you'd want to choose that by feel and understanding. They do have live data on board like on any other boat, I'd assume. And it makes a lot of sense for the sailors to know exactly when they want to initiate a manouver.
@michaellynch9284 Agreed. The interesting thing will be to see how far you can get with this kind of automation and where the sailors have to take over. E.g. I assume that the rudder and potentially the sheets cannot be controlled automatically. And even if they could, I assume a human expert would still be better at trimming the sails during a manouver than any preset script without any information on the conditions.
The second the machine gets all the measurement data and a few cameras we're in a completely different situation.
What are the limits on using a database to predict two seconds into the future based on the previous 3:00 to 0:02’?
Also, you can think of some of the automation like and engine map in F1 or motor racing, like how the throttle pedal responds given the fact that you’re speeding up after a tack or are likely preparing for one.
American AC teams constantly cry when they don't believe they have the technological advantage over others.
Its not fair unless they're the ones cheating.
I mean that's the modus operandi of USA in general, not just sailing.
I liked the introduction to foiling when it was introduced in the 34th... but all of the automation has just turned it into a battle of programmers and not really sailing anymore.... the only sensors they should have are the wind data, and location on course, and let the sailors figure out the rest on their own.
The 2 second delay can be overcome by an "autopilot" if it has fuzzy logic that can learn to use the time history, including the delay, to extrapolate forward into real time or the future. This will require the aquisition of vast amounts of data for the logic to use to learn from, but is very possible with a sufficient degree of accuracy to provide a base level of boat control, to which the crew can monitor and refine the final response in real time.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen (all one design supplied equipment). The crew reads that screen and can then input signals (press buttons, turn knobs) at will.Any fancy logic has to happen downstream of the crew input commands.
That could be problematic. The data can be re-acquired via a camera, and the availalble measurement data from the vessel systems can be used as additional input to guide the extrapolation, but it is certainly less than ideal.
The Kiwis would have looked at it long before the Americans. They're sharp as a 'tack'.
They've also written the rules so could have 'looked at it' years ago. I've no doubt they start with innovations already in their locker and write the rules to allow them without making it obvious to other teams.
@@TVRV8 I think this is the conspirator element of the story. ETNZ and INEOS wrote the rules. Did they use that power to to conspire to include an area only they had the resources to exploit? That seems to be the inference behind what Terry Hutchinson is saying
@@MozzySails The simple fact he's suggesting at it makes you KNOW he would have done it IF he could have so .... All is well that ends well.
Instead of "complaining" and suggesting they should be working on making their boats go faster.
Leave politics ashore and to the politicians.
What do you mean with the 'follow the dot' thing regarding ETNZ in 2017? Couldn't find anything online about it...
Welcome to the world of America's Cup conspiracy theories 😁
Not sure if it was confirmed, bit was certainly rumoured that Blair Tuke was tapping a dot on a screen whilst aboard his bike. Tapping the dot would break the 'computor autopilot' loop adding a human step where taping the dot would tell the computer to move the foil rake to the desired angle of attack.
There’s heaps of footage from 2017 with Tuke “tapping the dot”. I only saw more of it as ETNZ went deep into the AC.
@@MozzySails maybe AM buttons are actually mislabled to throw you off and maybe the dot tapping is also just a feint 🤔
The ‘follow the dot system’ was rumoured to have been developed by LR prior to them pulling out of Bermuda
Could you not train a model to predict what the wind speed will be in 2 seconds time? I'm not sure whether that could be at all accurate. Or perhaps you could have the sailors providing real-time input to guide the predictions. How long do these procedures take? Perhaps the model predictions are good enough to start the procedure and the sailors finish it off?
Exactly my thought. A derivative model using a continuous stream of data should be able to do that. Algorithms based mathematical models such as the Black-Scholes used in options pricing should be able to give you at least some advantage over flight specialist intuition.
Yes you can. I am coming from a Wave energy converter tech background. In this field, to extract the most energy from the waves, you need to control the force apllied by a Power take off (link to generator in most cases) to the prime mover, to control the system dynamic. Control theory gets better if you can provide your system with future wave loads, i.e. if you can predict what will be the loads in 5, 10, 15s. It is something you cannot do analytically, but with the help of machine learning, you can definitely predict the wave loads at a 10 to 15s horizon using all your device state parameter. So 2s delay will still give you some good level of forcasting. I am sure the application is different, but in an environment with waves with mean periods >5s, gust also lasting in seconds, it might be doable to train a system giving you more than 2s forecasting, therefore it might help anticipate some controls on the yacht, and not be reactive only
In vehicle ECUs the industry uses look up tables which is logged and stored from real time data (normally used as a fall back). This can be pre gathered and rendered down. a 2 second delay will allow you to get into a ball park, with this you can use well known deterministic algyrythms to best guess the next set of numbers.
This logic in machine control systems was developed as standard practice in the 90s. These (2010+) days everything in controller world is more sensor driven with target and actual values. EG you have a target and your closed loop system is reading the actual, it then has to work out how to get the system you're automating to the target value.
If sensors arent allowed in realtime .. the near realtime could still be used with prediction to calculate the best guess value and replace that missing data with its best guess (guided by the rendered lookup table from previous data).
Its what I would do anyways if I wanted to build a system like this within the ruleset.
Thinking out loud here, but if you had a sophisticated enough state estimator (math model) of the boat and wind conditions based on IMU or other legal sensor data, you could use it to output a 2s-delayed AWS value that the crew then compares to the telemetry data available. Then you just get them to adjust the up/down knob based on how accurate the estimator is travelling compared with the telemetry to give the model enough feedback for the Kalman filter jigger inside the box to do its thing.
Hey I don't think 2 seconds are a lot when considering the mass and inertia of an AC75.
And additionally, as many others pointed out, 2s delayed information is still amazingly useful for a computer that knows perfectly the behaviour of the boat and has a complete history of the status.
In my opinion once you allow electronics you are guaranteeing the ability to design an autopilot.
Maybe that is very advantageous to the defender, making the learning, designing and cost curves much steeper for the challengers
You can probably do this with a trained AI model. There is a natural rhythm of the boat, even across the two second delay, that can be understood and leveraged by an AI model enough to improve the handling of the boat, particularly if you capture enough training data from human control during the testing phase. This would be an interesting discussion to open, and probably worth a video on the adoption of AI technology in sailing.
Ahh, an American casting aspersions at Team NZ innovation, takes me right back to Fremantle!
Or san fran when yanks use automated stabilisation
@@MikeNEVILLE-xf4bc I try not to think about San Francisco 🤣
@@MikeNEVILLE-xf4bc many called it illegal
nope, he's talking about automatic operations that are carried out in sequence, with a pre-defined delay. This delay(s) could be adjusted on parameters that are not boat external (actuators pressure and position, ecc...)
I’d think of it less as a conspiracy theory and more like a fact. Surely they can script maneuvers and setup automations that would be useful despite a 2 second delays. Not to full on autopilot levels hence “quasi”
If you could identify a set of specific operations and conditions, like tacking in a reducing wind, or dipping another boat, or pinching to reach the mark, for example, you could use machine learning to collect data and associate it with these events. Then you could potentially optimize either the automated interaction between internal systems (foils to mast rotation, for example), or even the sensitivity or linearity of the manual controls, to improve effectiveness for future examples of similar events. Does the rule prescribe limits to on-board computing power and alteration of behavior during a race?
I'm not sure I'd call this an autopilot, but it is a refinement of the interaction coefficients between different controls on the boat, which could perhaps increase the performance of the boat over the course of a race or a series. This could be especially useful in outlier conditions, but it comes with the risk that the boat's behavior might not match the expectations and experience of the sailors, and therefore be counterproductive. Used with care I could imagine this being useful.
I think it would be possible to use the data with the delay in some situations, say trim optimisation or small adjustments on rudder position to head the boat up into a lift. The reason that I think it would be possible is that they would not really look at the absolute data, what I think would be more useful is looking at the trend in data that humans don’t really have the ability to use. If you think about when you are helming, you have a variety of inputs, you can see the colour of the approaching water and as you get to the edge of a wind band you can make significant adjustments to trim or rudder input to take advantage of the change in wind speed and direction, but it may be that if you look at the data, the transition isn’t as sudden as you think and there are some clues that you can see from the data if you have a richer data set. To some extent as a helm or crew of any performance dinghy there is a lot going on and I expect that everyone becomes overwhelmed with information and inputs, but if you are an electronic device that doesn’t have to listen to the other person in the boat yelling at you about why you called a lift when it wasn’t or suddenly having a face full of water as you crash through a wave, then you can have a much more objective review of the data. Anyway, I’m not in automation design for americas cup and I’ve not actually looked at the data the external facing systems produce that’s delayed in that way.
Hi,
Are boat pitch sensors ( = foil flaps control to keep stability and avoid water touch downs ) also delayed 2 seconds?
thnak you !😉
exciting. I’ll also spray some cologne on the fan. No one prohibited visual communication with the sensor. It is enough to place a millimeter scale (sorry, inch) next to the direction sensor. Next to the speed sensor, a tuning fork. and view them via gopro. the speed of light is the maximum speed possible in this space.
Outstanding content 🙏
Thank you 🙌
Thanks for the insightful video Mozzy!
Could you elaborate on the "follow the dot" mechanism etnz had in 2017?
Sensors showed the controller where to move the ride height, dot on screen moved and controller mover joy stick to follow it. Clever, at least it was not cheating like Oracle did in San Fran with Herbie.
Having ANY electronics control doesn't do it for me. I would also like the cup to be ran regardless of wave height. If a team and boat is better then it shouldn't need precise conditions in order to run/compete.
You don't need a computer control to link mast rotation with foil position, that task can be performed with a mechanical linkage.
So often with sophisticated techniques it is the use of (probably hidden) layers that are outside the comprehension of those without due expertise. We as outsiders may be stuck in mindset of a single layer that does it all, rather than the interaction of multiple layers each doing a smaller thing with predictability with all scenarios mapped so as to avoid the unforeseen. Reduce the cognitive load on those pushing the buttons and seek some small advantage. What weaknesses of the boats might they be seeking to target because those limitations may more fundamentally create the character of these vessels.
If all that is required is for the mast to rotate when, or just after, the foil is dropped or raised then no feedback is required at all. It's just one switch doing 2 functions which is not difficult to achieve - with or without a computer. It's barely even automation.
If I flick switch A then the port foil drops. If I flick switch B then the mast rotates. Or if I flick switch C the foil drops and the mast rotates. Where is this claimed non-permitted feedback?
There is no claim of evidence of non-permitted feedback. I just showed the mast as evidence of what was allowed and think it's a pretty basic relationship and now really that much more sophisticated than linked controls on dinghies (spinnaker goes up and pole goes out on same line).
The claim that something else can be done is purely speculation from Terry Hutchinson. I am just outline the technical context for it and am sceptical myself it is possible to achieve.
Great vid as always. I think it is a good question to ask for every control/adjustments on the boat, can a human see something and react in significantly less than 2s rather than the computer seeing and reacting with a 2 sec handicap. I suspect there are many controls that a human can’t react significantly faster and so thereby gives both a cognitive advantage to the humans by load reduction and/or more precise changes where automation can do better.
Correction...the Oracle system did NOT include external data in the feed-back loop. It was manually controlled
Excellent thank you, great to know about mast turning automation etc, I'm as excited as you are to see the new boats, flights and accom are booked 12 - 20 Oct can't wait, go ETNZ
No surprise that team NZ set up the rules in a way they can game them. The automation will spell the death of the event as a sporting spectacle that's supposed to be all about sailing skill.
The boats are big with a lot of mass. 2 seconds can be productively utilised I'm sure. There will be plenty of of lead/lag power relationships on board. Being able to dynamically manually adjust those relationships(AWS +,-?) at different speeds would be key though, i.e. powering up the mast angle 2 seconds after the middle of a tack & balanced foil cantilever angles. As a kiwi myself, I have to say that it is in the nature of the America's cup rules and history to use boat design rules tactically against the opposition. Were it me designing the control algorithms by the permitted means, whether fluidics, silicon logic or other, the goal is to put the hydraulic operation sequences into a natural harmony. We have history in other fields, for example Fords when their designers were still learning how to write software, of loosing inertia for flywheel mimum RPM retention under light injection and intentionally dropped cylinder's combustion sequences(to maintain some stoichiomeric mix in active cylinders at partial loadings). They fixed that with crank angle sensors and presumably a SW update. I knew because I'd invented my own engine with the same design goal practical implementation concept. To put those sorts of factors under manual control is a bad idea, you have high incident repetition with tight timing, it's not well suited to human control(the Ford would eratically stall on take-off). So I welcome this sort of rule change. It's forward thinking and analysis like you are doing is potentially very rewarding. I think the means of implementation of the logic is irrelevent, that horse has long bolted the stables. Those who criticise these high speed foiling works of advanced engineering teams yearn for simplicity and low IQ requirements gear but that's not what this cup racing is about.
i've always said this inconsistency in contemporary AC racing. There is no checks on the software load. It could be full of cheats.
One fix: Have a configured software load that goes into the boat right before the race. Teams must submit their source code in advance of the race.
Shades of Denis Conner wanting core samples😂
If you have systems that can react fast enough, could you theoretically use sensors on the jib's control systems as a pseudo wind vane to automate certain aspects of mainsail trim? Since these boats are basically always sailing upwind, the jib will always experience changes in wind angle and speed slightly before the main does. I would imagine that these "3D sheeting systems" that some of the teams are using on their jibs would allow for more accurate data collection from the jib itself due to the forces being distributed across multiple lines that could each have their own load sensors on them.
Aye! The people who write the rules usually have an advantage.
But it's the people who *understand* the rules best who usually win.
I think that he’s on a fishing expedition.
There’s a vast difference between automation of linked systems functionality to reduce crew workload (improving safety by reducing critical systems operating c^ck-ups) and an autopilot that will get you from point A to point B without human input.
The 2 second delay would effectively render any such implementation less efficient at the task than the human crew, doubtless why that stipulation was made.
Having spent time as an engineer in motorsports during the transition period from carburettors to fuel injection, I can hear echoes of a number of objections that did the rounds at that time.
They were the exclusive preserve of older and more dogmatic individuals who were, quite frankly, out of their depth in terms of how the technology functioned and what it’s capabilities were.
After that frustrating experience I’m just glad that I’m not even remotely involved in attempting to police these regulations, or indeed to defend the innocent against one unfounded accusation after another.
Not that familiar with the Americas Cup control systems or rules but when we did your foiling electric boat there is definitely a way to use the 2 second delay information to tune the system. We would run test and simulation systems using hardware in the loop and then use the data coming out to then modify the gains and derivatives on the controller to improve the flight response to conditions and optimize. So if this all they are doing on the water then your comment @mozzysails would make sense that they had buttons with + and - for various variables as the data comes through they can then make the changes based on the new info coming in even if its 2 seconds delayed. One of the best tools we used was to have the ability to plot data and then use that to see if the changes were moving in the right direction or not, so by overlying the right data you can see if you are making efficient decisions. So in essence its not automation but more semi manual manipulation of the system to give better performance. The 2 second delay could also be giving feedback on how much your settings were out and from this you can then know how much your "feel" for the boat is going, so for example you trim the foils and them 2 seconds later you know what the performance output was. When you get used to this method it almost becomes "normal" as opposed to being delayed.
One thing to consider is that TNZ utilised AI to drive their simulator in the last cup. It wouldn't take much for the internal systems to now have an AI component in charge of the automation. You could train the AI in the simulator and then upload the latest trained AI to the boat. It would be like having an additional artificial crew member.
Can you make a video on the "follow the dot" program by New Zealand in 2017? How did it work? I searched for info, but only find short mentions of it online. What is known?
It's getting spicy! I love it. Thanks Mozzy.
You bet
You have given my brain a
Thanks, will do!
2 second delay is perfect for working out the sea state. If you use cap sensors on your rudder you can get mm accuracy at what ever bus speed you can read. Much better than ultrasonic which is like 10x per second cm accuracy.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen. The crew reads that screen and can then input signals (press buttons, turn knobs) at will. So the control circuit which can link controls can't be running on actual data with fancy algorithms.
@@MozzySails yes but can they process that data first? if they can, they could use that to establish the sea state, wave heights, periods and also if your on a lift or a knock if there's mcu data. The thing with cap sensors you can embed them in the hull. So if you can use that data even on a 2 second you can establish targets that can then be used with the allowed live data from load cells and do a chase the dots. (I made a flight controller and was switching to use cap sensors when covid hit and I gave up as a lost cause)
@@andrewferguson8735 I don't believe so. If you pause the video at 7:19 you will see in the top right of the diagram that everything in grey is a supplied system that is the same for all teams. That allows data to appear in the Crew Information System (CIS) and display on the crew indication devices.
The only thing I can think of is that the crew induction device could process the data and display recommended settings to the sailors.
You will see there is no link from CIS to the Electronic Control Systems during racing, so any commands that go in to the ECC must come either from the crews input devices (buttons etc) or from sensors within the ECC.
I think if an AC team really wants to put doubt and a fright into the other teams, they should hire the former CEO and the engineers of VW who masqued the diesel software emissions controls to dupe the regulators. Could an AC team rig their electronics to fool the Official AC boat measures? Maybe. 😃
Bring back the 12M with no computers on-board. When sailors were sailing and tactic was the game.
Agreed , bring back wooden hulls cotton sails, no life lines open cockpits and a proper skipper wearing a peaked cap and a rolled neck sweater, smoking a pipe of ready rubbed St Bruno tobacco.
Every boat should be free of advertising and named after a sea creature or exotic wind events.
Bare breasted carvings on the prow hand painted by students of the Salvador Dali school of art.
Oh last thing, no black lesbian trans volunteers to be allowed a place on the crew without a note from their mum. 😅
@@skippyone3085 take your med, mate.
Exciting stuff, but technology has really levelled the playing field and to be honest, the racing gets a little tedious. Think I preferred it when it was more reliant on manual skills
Excellent comment Mozzy!
Good question. Am from newzealand do think newzealand has had this innovation on point from the get go am thinking they would have worked it back to stay legal will at least hope they did .they will have lots of options that will be available questionable to say the least. Guess they're waiting to see if needed. As will have to explain if do .
2 second delay is ok if they are also allowed to run predictive algorithms. It's like weather forecasting in microcosm, no? You just compute the differentials, and run the dynamical system (or these days incorporate a neural net for some of the number crunching). Means they have to gather a lot of data at the venue _beforehand_ to train the ANN in advance. You'd have to prove me wrong to discount this. A sailor "looking well ahead" can probably be beaten by a decent ANN if the training data match the current weather well enough. Might only be a minuscule advantage, but... this is the AC.
So here's the thing. The data displays are supplied equipment. So as I see it, it's a one design link from measurement devices for environmental factors (wind, speed, ride height etc.) with a 2 second delay. The crew then reads from that screen and presses buttons. So in terms of fancy linking of controls, that all has to be done in the electronic control circuit downstream of the crew inputs.
I could see if you could get the environmental data in to the control circuit, then have some fancy algorithms which forecast future best responses, then you may get something that works even with the delay. But I don't think this is possible with the 'human link'.
It's crazy to have stuff in the rules that the event officials can't see.
Great update. Automatic sail and rig trims seems sensible, but any update that brings "less fun" will surely end up being self defeating. Porsche 911 in manual gearbox spec shows what the market thinks here.
Also, Is the term "gybe" now superfluous and unnecessary? Given awa, why are we not just tacking downwind?
No, that’s not sailing if sail and rigging are automatic, that’s part of a skill set a sailor should have
Very complicated boats, and it seems that this means there are going to be grey areas. For instance they need to have control systems for the ride height that react faster and better than any human. Is that considered a kind of autopilot or not? We know that there is far too much data for the helmsman and strategist to respond to, so automation has to come in otherwise these boats are unsailable. Another complication is that once the computers have been set up, it is virtually impossible to read the code back to figure out what exactly it does. Will the organisers have to scrutinise the code on every boat to ensure that there are no transgressions into sailing autopilot technologies. I'm glad I'm not a race organiser.
I'm sure as a AC skipper, what they would love is a boat that largely sails itself, barring a few nudges on the wheel and a couple of button presses per tack, so that they can focus out of the boat on what's happening tactically on the race track. Is this we want out of a Yacht Race? When I started sailing, all I had to coordinate were two ropes, centreboard height, body position and the telltale in conjunction with reading the wind, waves and current. It was very visceral and physical, and mostly quite slow. I do enjoy watching these Formula One boats racing, I like innovation but I don't want to see cheating.
Anyone have a link about the TNZ "Follow-the-dot" system?
read this: www.sailingscuttlebutt.com/2017/06/19/americas-cup-stick-a-fork-in-it/
I thought they said that ALL THOSE BOATS ARE THE EXACT SAME?
I thought that anything above the waterline had to be powered by human force, while anything below could be battery powered. If that is the case, how can the mast rotation be automatically synchronized with the battery powered movement of the foil arms?
you're correct about the power sources, but signals to both go through the same electronic control system so can be linked there. There is a degree of stored power in the human system vi hydraulic accumulators, but essentially the valve opens to turn the mast and as long as the cyclors have built up enough pressure for it to move, then it will move.
Thanks for your explanation.
Americas Cup is no longer for sailors, but for engineers and programmers... The spirit has been corrupted, engulfed by the need of win for the sponsors.
Don't get me wrong, development is always a must, but I would not allow a single piece of electronics onboard. A handheld radio and that's it.
What’s you opinion about the new switch one design?
Can you do a video about it
yes, planning on getting down to Weymouth to have a go. Looks cool. I like the concept of one design which freeze in time a peak from a development class
And here I am sailing a boat designed in the 70s built in the 80s that spends most of its time being steered by a mechanical system that was pioneered in the 30s (Moody 37, with wind vane steering)
She ain’t fast, and she doesn’t keep to the rhumb line but it usually averages out pretty good, and she’ll get you there… eventually…
The crew become 200ms collision & weather sensors, along with power generation for the computer that works for a bigger computer.
There should be no automation. The sailors should have to control everything.
Perfect information from 2 seconds ago plus an amazing autopilot might be better than human perception in real time with human input.
You are assuming that the data they're collecting is raw and from a sensor feed or array of sensors.
Most likely it would be from the inputs made by the crew, in the form of certain conditions, like Wind speed, angle of attack, current sail and daggerboard conditions, etc.
In that case it would not have to be a 2 second delay.
Not necessarily an Autopilot, but a “Sailing Assist”
The 2 second delay at the speeds these “boats” are going changes the evaluation. There should be NOT any digital connection or “automation” or adjusting or movement of “systems” for sailing
What if there is a 2s delay from sensor to execution system, but there is a machine learning predictor logic, trying to “learn the future” and having an instant (2 second) feedback loop. And the execution system makes decisions on the basis of the prediction.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen (all one design supplied equipment). The crew reads that screen and can then input signals (press buttons, turn knobs) at will. Any fancy logic has to happen downstream of the crew input commands.
Is it an autopilot if it shows you the optimum setup and a human follows that setup? The two second delay may be the delay from switching from one mode to another. For instance foil trim and then sail trim/mast position. In the video showing the foil dropping and the mast turning, there's a two second delay between the actions. Is that the loophole that Team New Zealand are exploiting?
Memories of Dennis Conner, preparing excuses before he lost.
6:41 Occam's Razor = The 𝑠𝑖𝑚𝑝𝑙𝑒𝑠𝑡 explanation is the most likely.
High quality analysis as always. Thank you. One thought - is there any way AI could learn to predict the direction of change from a two second delayed feed. If I saw the same thing happen ten thousand times followed by a predictable change, couldn't I effectively eliminate the delay with predictive logic?
I think you could run AI on it during training and in the simulator. Some sort of machine learning really that can then produce algorithms to be used in the racing to coordinate control... but I still think the 2 second delay would make it really hard
I wonder if the idea is that computers could be used to "simulate" what would be ahead on the race course, essentially predict the present and use that for a quasi autopilot
As always, great video !!
Could you do a video explaining the "follow the dot" system that ETNZ were using in Bermuda ?
There isn't a lot on it, despite what felt like lots of discussion at the time. You can read this article on it here: www.sailingscuttlebutt.com/2017/06/19/americas-cup-stick-a-fork-in-it/
Good video! I could see teams using AI to fill in the 2 seconds 😮
I think it's a lost cause disallowing autopilots but allowing "blind" autopilots. What you will get is crew control becoming the environmental input for a neural net. There will be no end of accusations and counter-accusations when it comes to that input into the autopilot. One would like to turn back the clock and disallow feedback control, but that ship has sailed, pardon the pun.
3:20 "It's pretty clear irrefutable evidence"
It's not though, it's not even bad evidence, it's just a video that shows the boat doing two things. it's _you_ who insists that the mast cannot rotate by any other means than automation 'because why would anybody rotate the mast while not moving?', except you just explained that lowering a foil is _always_ followed by rotating the mast so why is it now suddenly strange that they did that here? It's what they train for...
What you see in the video literally is the routine they go through when lowering a foil, always, on every boat.
It would be evidence if the guy in the cockpit tells us all he did was push a button to lower the foil and then the mast rotated, but that's not in the video.
And even then we can get into all kinds of nooks and crannies about things like having a button that has two contacts; one to lower a foil and a second to initiate a mast-rotation after a five second delay, which would give the same result but is not automation.
Not saying they do this, obviously, but my argument is that the video is does not show what activated the mast rotation.
Great video as always! But talking about Luna Rossa's control buttons, you mean TWS+ and TWS- meaning true wind speed up or down, right?
yeah, going from memory of what I saw last year! Realised it in the edit, probably should have put a note in. I think the principle still holds.
They also have a UW/DW button which could be to switch configuration for each leg. Then the green and red button pressed for each tack to rotate mast and raise board.
the amount of sailing and wind data these guys have and the data analysis capabilities available to them makes me wonder if theyve been able to find very fine and small patterns in the wind around barcelona, and then use these patterns to be able to predict 2 seconds into the future?
maybe a coarse autopilot is available from this automation rule and processing power? something like steer assist in a car racing game like forsa or similar.
its a wild thought.
Guess they can turn it off during pre-start or roundings with both boards down?
but what if you have the human subjectively evaluating the wind speed, looking at gusts etc, not looking at delayed sensor data but looking at waves etc. not accurate you'd say but if you spend time training a number of professionals and selecting the best ones at it I reckon it might be quite accurate?
I don't even understand why we're still able to follow this competition!... What happened to the sail changes, the gybes under spinnaker, the luff starts and everything else that made this competition so beautiful?
Today, it's just six guys pumped up on protein, grinding away non-stop and 'water snatchers' speeding along at 40 knots on a lake.... We're bored!!!
With modern AI it might be possible to train a neural net to forecast the values based on the current ones. To try to minimise the 2 second delay in the data.
the thing is I don't think you can get the 2s delayed data out. It only displays on a screen (all one design supplied equipment). The crew reads that screen and can then input signals (press buttons, turn knobs) at will. Any fancy logic has to happen downstream of the crew input commands.
6:40 Occam's razor does not apply to Formula 1, NASCAR, OR America's Cup.
Very true!
Possibly the autopilot won't so much be an autopilot but a man in the loop synthetic control system. In the airbus for instance the pilot makes control inputs but the aircraft then auto trims to hold the aircraft in that new attitude. Perhaps they will be able to reduce the load on the trimmers. Always running 2 seconds behind does seem a problem. Possibly it would be more of a back up for when a sailor hasn't had time to do something yet.
As a person just watching this as a sailing fan, this is very dissapointing to learn.
They should ban electronics from the AC, and make it about sailing and sportmanship again.
I think it’s totally doable. If I were working there, I would implement Kalman filters to predict all the environment information at present given them at 2 sec in past.
Straight line speed optimization is totally doable here. Esp. With a sailboat and a known 2sec delay.
It’s not rocket science, it’s known robotic science.
I suggest that in the Formula One world there is an underground river about those themes and the battle for win is decided day after day more by electronic and sensors and AI than an engine or driver ability
Data logs fine tune the boat, looking at data and adjusting the boats tune makes them faster in certain conditions, just like car racing. ecu tuning, different tires and pressure for different tracks/conditions. car doesnt know how fast the winds blowing or which way . the driver has to keep the car straight and adjust accordingly, speed, steering wheel etc.