AI FTP Detection Update

[enki42]

@asteryx

Also I’d like as a user to decide whether to prioritize FTP up or PL progression out.

I don’t think this would be useful. Sure, pick the energy system you want to emphasize or endurance vs sprint so the workout target the improvements you want but the point of the AI is to figure out the best way to target that, not you telling the AI what to do

[toyman]

Intervals.icu and WKO do their FTP modeling based on the power/duration curve, and require maximal efforts of different durations for that model to be accurate. WKO even provides graphs that show what time durations you likely don’t have maximal efforts for.
To me the most interesting part of TR’s work in this area is to estimate your FTP without an accurate PD curve. If you are following a plan, you likely only have maximal efforts for short durations - VO2max intervals and shorter, and even those may not really be maximal.

[Nate_Pearson]

@Helvellyn

It think it does make sense, in that whenever your FTP is reset your PLs change, and so the AI will say "Aha! You had worked your way up to 3 × 20 minutes Threshold workouts, but there’s no way you can handle that at this new FTP, so I’ll knock you down to 3 × 12s.

Not what you want if you’re trying to work on your ability to work hard for longer. Out, then up. Out, then up.

Yup! You got it!

And I suspect that if you’re new you want to go Up then Out. So if your first FTP is 140, it’s probably better to get that up to 210 (just an example) rather than focus on holding 140 for 60 minutes.

[Nate_Pearson]

@Grundy82

I guess the acid test would be how sensitive in predicting change the model is. Too sensitive and those frequent changes may become demoralising for some users - I could see this particularly for those who may not fully understand the change is designed to ensure optimal training and time in ‘zone’ / PL optimisation (be that a change up or down in FTP).

Yup, we understand that it can be too sensitive and are making sure that doesn’t happen.

@Grundy82

So, if I’m understanding this right (assuming the released feature covers Auto & Manual), the user would have the option to ‘toggle’ between either ‘Manual’ or ‘Auto’; if in ‘Manual’ mode, the Ramp test workouts would still appear - as it does today if using AI FTP Detection - but the user could effectively choose to either a) accept the AI FTP (and get a replaced workout) or b) conduct the ramp test? On the flip side, if in ‘Auto’ mode you’d never see the ramp test, it would always be a workout by default and FTP would just be auto set by AI?

Would the auto detection differ in the user message subject to whether that’s being set based on the start of a plan vs mid plan i.e. as a replacement for the ramp test vs how you’re performing / progressing mid plan?

Not exactly. Here’s a better summary of the vision:

AI FTP Detection is run after each workout. It’s on for everyone and you’d see it in a summary screen along with your PLs after a workout. This is a smart feature that will take many things into context (including plan phase and PLs) on when to change your FTP.

For the “power users”, they have the option to run a manual check every 4 weeks. We don’t want to encourage this behavior but we’ll have it as an option. Many athletes won’t even discover this.

Plans will default to not having ramp tests but have an option to add them. For those on a plan with ramp tests, you could get a detection on that day (if it’s been >28 days) and have the ramp test switched out for a workout.

[Nate_Pearson]

@RobertSims

Love this idea…but for small changes in FTP, rather not bother… let’s face it - if your FTP doesn’t end in a 5 or 0, your kidding yourself about accuracy!

I.e. rounding to nearest 5 watts is accurate enough (especially with error % of power meters).

Don’t worry, we won’t be presenting 1-2 watt changes. It’s better just to work through your PLs.

[Nate_Pearson]

@enki42

I think the better question is if ai ftp your current ability or what your current maximum should be? Current ability means if it’s lower for any reason, it will be lower (fatigue, feeling sick, break from training) so needing more recovery means you’re ftp could be going down. So kind of need the red/green functionality so it can give better context is saying it detects you are fatigued so will decrease your ftp and tell you to rest more. Otherwise you could see it as a reason to push harder and dig yourself deeper

Think of it like this:
FTP = Benchmark of where you are.

Progression Levels = Fine-tune adjustment of different energy systems. Effectively remove the relation between VO2/Anaerobic/Endurance/etc with FTP. If you get sick PLs will get lower (depending on how long you are sick) and will zero into your fitness post-sickness pretty quickly.

Red light/Green light = Try to protect the athlete from doing too much. IE doing a big group ride on Sat then a race on Sunday and a Vo2 on Monday. Then they fail Monday’s workout and say TR is too hard.

[Nate_Pearson]

@redlude97

wouldn’t the correct solution to be to specify the new 3x20 progression at a lower % of the new FTP and work up from there? I feel like there is still a mismatch in terms of PL fitting with plan with FTP detection. Especially if the change in FTP detected is small since it is being assessed after every workout. Shouldn’t need such a huge drop in TiZ

It depends. If you’re going to do a 40k TT or in another specialty phase, you probably don’t want the bump in FTP.

But if you’re early on and have many months until your A race, you probably do want the bump and you’ll end up in a better place in the end.

I also think there’s going to be a correlation between watt/kg and experience in this but we won’t have that feature in launch (but we’ll definitely look at it).

IE I don’t agree with anyone who says you ALWAYS want to extend TiZ or you ALWAYS want to push up FTP as fast as possible and just keep doing the exact same workout at different FTPs.

[berto2cj]

Only thing that jumps out is completely removing tests from plans and calendars…keep it there - run auto and swap it out like it happens today. or maybe an annotation,…

to me removing all reminders of it seems too far to jump right now.

PS - I’ve not tested since auto came out…or when I did I gave up and listed to AI

[mcneese.chad]

@berto2cj

to me removing all reminders of it seems too far to jump right now.

• But why should they keep testing in place if the Auto setup will handle the changes?

I am just trying to understand what you see as the specific benefit of having them, or the detriment to removing them.

[berto2cj]

Revising / rewording

Some visual calendar indicator where i can see future instances of auto FTP that are planned. I get it could detect and adjust sooner.

i’d miss weekly blocks - that’s all

[Helvellyn]

That’s spooky. I just got an FTP bump to … 210.

Please could you give me 6 numbers between 1 and 49?

[mailman]

So, this will be out Monday right?

[SteveITKC]

So this is all very well and good but without a solution to unstructured workouts, I really wonder how accurate an AI FTP detection is in whatever form all these interesting options suggest it ultimately takes.

While I’m patient, I am disappointed to see all the attention that you/ @Nate_Pearson give to something like this and the relative lack (absence?) of official/public TR attention you give to unstructured workouts in the forum. I check the forum and podcast each week to hear if/when it’s happening or any progress only to very occasionally hear @Nate_Pearson slap down @IvyAudrain or @ambermalika for mentioning it and purposely being vague about when it will be out.

It’s summer, we’re riding outside and unstructured workouts is a thing. Would be great if you can update us and show this feature some love too. Thanks.

[ChrisDe]

My understanding was that the PL were a “workaround” to the two main problems with FTP coming from the ramp tests.
1 - FTP is not up-to-date. The further you are from the ramp test, the more the FTP has changed, so PL need to increase to keep the workouts challenging.
2 - FTP does not take into account your power curve profile. People with the same ftp can have very different anaerobic capacities, and therefore very different VO2Max PL.
Now that the AI FTP detection has been developed, these two issues could be addressed. The FTP could be updated after each ride, remaining correct all year long. With a little tweak, I’m confident that TR would also be able to provide the correct power needed for the other training zones.
I understand the need to extend time in zone instead of increasing the power, but I would rather have that linked to the timing in my training plan rather than linked to the time my FTP has been updated (I didn’t read anywhere that the auto detection would happen only at the start of the plan).
If I’m in week 1, I should do 2x10min, week 2 2x12min etc, with the AI choosing which power to do during the exercise. This would be the same idea for vo2 etc
How does it currently work for people who are plateauing? Do PL get stucked and they get no progression?

[grwoolf]

@Nate_Pearson

This is a smart feature that will take many things into context (including plan phase and PLs) on when to change your FTP.

For the “power users”, they have the option to run a manual check every 4 weeks. We don’t want to encourage this behavior but we’ll have it as an option. Many athletes won’t even discover this.

@Nate_Pearson

It depends. If you’re going to do a 40k TT or in another specialty phase, you probably don’t want the bump in FTP.

But if you’re early on and have many months until your A race, you probably do want the bump and you’ll end up in a better place in the end.

I also think there’s going to be a correlation between watt/kg and experience in this but we won’t have that feature in launch (but we’ll definitely look at it).

IE I don’t agree with anyone who says you ALWAYS want to extend TiZ or you ALWAYS want to push up FTP as fast as possible and just keep doing the exact same workout at different FTPs.

First of all, I’m really impressed with what I’ve seen from AIFTP so far. I’m pretty in-tune with what my FTP should be (at least as a basis for dictating workouts) and the AIFTP recommendations have been within in a few watts of what I would have set manually. It sounds like it’s working well for others as well, so this is next level awesome stuff. Really nice job and congrats to the folks designing and developing this stuff.

I particularly appreciate that AIFTP seems to be decoupled with TR workouts and looks at structured and unstructured work. I’ve been using a very messed up combo of TR workouts w/ adaptive training combined with unstructured outdoor rides this year. Some of that is by design and some of it because I’m out of sync on my training plan and too lazy to adjust it (I just keep changing out workouts individually to fit the phase/week I really want to be in). It’s nice that AIFTP seems to ignore all the noise and just look at what I’ve done. I’m hoping progression levels and all of AT will eventually get there as well.

While I don’t expect TR to optimize a training system around any “non-standard” approach, I have the idea in my head that the system should be primarily focused on what work was actually performed and not care much about what was planned. I totally get the argument that auto FTP might want to look at the phase of training as a data point, but for me that would be a much less relevant consideration compared to work performed. I’d have the same concern with progression levels being considered, particularly until progression levels are accounting for unstructured workouts. All that said, these features/solutions currently being developed are really ambitious and should be applauded even when the initial releases might be limited. I’m guessing there are some really smart folks figuring out how to release these incremental features while keeping an eye to the long term vision. Again, kudos to the team and I’m looking forward to what’s coming.

Lots of rambling there beyond Auto/Manual discussion, but I think the Auto FTP thing should be the primary solution. Bury the manual option somewhere (it’s sort of buried today I’d argue). I personally don’t like the 28day fence restriction, but I get the argument for it. An argument/situation for a shorter re-assessment - I have an abbreviated 2 week build block that I want to squeeze in for whatever reason. A couple weeks of “new” intensity following a big volume base block could easily drive double digit FTP improvements. Maybe it’s enough to see the increased fitness in the higher progression levels, but I think the FTP number will continue to be a carrot that many folks want to see. There’s a strong argument that it shouldn’t matter, but FTP is too ingrained to ignore. Another situation where the 28 day limit hurts - A user gets curious about their FTP half way through their build cycle and does a manual calc. Then, 2 weeks later than want to see where they are at the end of build, but they can’t until half way through specialty because they have another 2 weeks before their waiting period is over. Constantly updating FTP is not ideal, so maybe there should be a limit, but 28 days seems long to me.

[yokuha]

@mcneese.chad

TR drops PL’s because it’s not necessarily likely that the rider can take the new FTP and perform the same workouts at the same PL’s. This is a practical solution to the reality & anxiety we have all seen or had after getting a higher FTP within a TR program. I and many others often struggled with the workouts following an FTP bump within the old TR plan setup.

Should really does this, as increasing the PLs on the old nominal FTP already compensated for the somewhat increasing real FTP…
Thus, when adjusting nominal FTP have to set PLs back to your-nominal-FTP-equals-your-real-FTP PLs.

[yokuha]

@SteveITKC

So this is all very well and good but without a solution to unstructured workouts,

Yes, it really needs to take “rides” into account. My best FTP estimates come from races, real or practice!
Both in the sense of higher as well as that I can train well to this high numbers. Thus, these rides should be taken into account both for eFTP as well as PLs.

Right now, I once in a while manually update my TR FTP from a new I.icu “max. effort” eFTP, but it would be great if TR’s ML would figure it out itself to have a truly consistent model.

[ellotheth]

@SteveITKC

So this is all very well and good but without a solution to unstructured workouts, I really wonder how accurate an AI FTP detection is in whatever form all these interesting options suggest it ultimately takes.

@yokuha

Yes, it really needs to take “rides” into account. My best FTP estimates come from races, real or practice!

AI FTP looks at all your rides with power. From the official FAQ:

You’re confusing AI FTP with Adaptive Training, generally. From the AT docs:

[yokuha]

@ellotheth

You’re confusing AI FTP with Adaptive Training, generally.

Indeed, I guess I do/did.
However, then I do have a problem with AI-FTP detection!
Multiple times this spring I’ve manually put eFTPs based on i.icu’s (power duration curve) analysis into TR, because the ramp-test-equivalent AI-eFTP from TR was significantly lower, e.g., >10 W lower, and seemed to low to me (and I did not want to perform the actual ramp test that Monday, see Combining TR plan with other training activities ;-).
The subsequent TR training/workouts went just fine and both FTP and race performance was good;-) I then also always quickly went up to PLs >5, sometimes up to 9, on the systems that the planned TR workouts focused on.

[PeteFranks]

Wow Auto would be amazing please go ahead with this feature guys