IW STi Forum banner

1 - 18 of 18 Posts

·
Registered
Joined
·
2,256 Posts
Discussion Starter #1
Hello all,

I've been modding and tuning my cars for about 10 years now and while the automotive industry as a whole has made a ton of progress, I feel the tuning industry has lagged behind, big time. It seems we're stuck between chip (or canned flash) tuning and human-written dyno tunes and both have their drawbacks, particularly in cost and accessibility for the latter, and precision for the former.

I find myself considering what could be done with modern cloud and machine-learning technology. So I thought I would come to some of the potential users of such a product and get some thoughts before I decide whether or not to pursue.

What I'm envisioning is a cloud application that takes data about your car via logs, your mods, and conditions as an input then runs it through a model which would be constantly updated with data collected from your own car and every other car on the network based on their relevance/similarity to your own, in the simplest terms.

The major benefit of this would be, you could put a new intake and header on your car tomorrow and instead of scheduling a tune, you would just select the modifications you added or input some details about them (like MAF diameter for an intake) and the system would get as close to a base-tune as possible, then make real-time adjustments to the tune via OBDII until it got as close as possible to perfect, the same way a pro tuner would in person. If your ambient conditions change, the tune would adjust appropriately based on all of the mapped tunes, and so on. So the more cars on the network, the better the tunes become. I imagine there would have to be human tuners on call for circumstances when the network can't identify the cause of some unforeseen signals from sensors or hardware issues.

Hardware would be an OBDII cable (or a bluetooth OBDII adapter if it can handle the bandwidth requirements) with an adapter for iOS & Android then a mobile app to input data and connect to the network.

Cost wise, I would think an initial fee for the hardware/app then a small monthly fee to continually access the network and get updated tunes.

What would be your concerns if you were looking at such a product/service? Would you be willing to pay an additional cost vs a one-time flash tune via a device like an accessport? Would you be hesitant to trust a new brand?
 

·
Premium Member
Joined
·
5,331 Posts
Interesting idea.

* It would be useful to have full time logs of any car and OEM is headed this way, no?

* I don't think it would be nearly as successful for tuning - except for fine tuning. Why? Cause good tuners get it right and except for fine tuning the tune should not change. Before or after cars have issues. All kinds. If your could diagnose those and inform them they have a bad turbo or head gasket that would be awesome. I don't think that would happen without a lot more sensors.

* how do you determine a car that has an issue from one that tuning might help, cause otherwise you are trying to tune a car with an issue, and unknowingly putting bad data in the data base and using it to tune others cars!

* When you first offer this what will you do about pro tuners hooking up cars and telling you they have very different cars than they actually do. Pranksters might do this too, but try to compete with others jobs and it's a given you'll have to deal with this . . .

But interested anyway? Sure!

ADDED: 8cd03gro - didn't notice this was you - not that it matters but I'll answer your other questions:

What would be your concerns if you were looking at such a product/service?
Concerns? same as having a black-box. What will you learn an be forced to release about me? Can't get around em in modern cars anyway.

Would you be willing to pay an additional cost vs a one-time flash tune via a device like an accessport?

Absolutely. If it worked well, "we all" would.

Would you be hesitant to trust a new brand?

I'd rather have a proven tune but you build a brands trust
 

·
Registered
Joined
·
342 Posts
i would be interested in the fact that it sounds like it would be adjusted to my driving style instead of a few WOT pulls or guessing with A/F or ET gauges...say i prefer to put 86 octane in my STI (i wouldn't do this) but it could adapt to keep me going...alternately if i want to F1 race to and from every destination it would do the same...I currently have a great dyno tune that is keeping me going and my ratios/temps have remained great even with adding more parts, i could benefit from doing a retune, sure, but scheduling it (4 hours seems to be the average locally) is hard with work, kids and all the life that goes with it...i agree on the blackbox theory, also adds concern of the potential for someone to hack my car, bad tune is loaded and engine is trashed. i think the AI could spot pranksters loading fake data as there would be no feedback from those vehicles to the system, much less data would be different streaming it, simple OBD2 code readers can read the VIN and thereby allow the system to verify vehicle data/engine type and other sensor data...monthly subscription costs, sure, if it works and does the job correctly...
 

·
Super Moderator
Joined
·
8,410 Posts
i would be interested in the fact that it sounds like it would be adjusted to my driving style instead of a few WOT pulls or guessing with A/F or ET gauges...say i prefer to put 86 octane in my STI (i wouldn't do this) but it could adapt to keep me going...alternately if i want to F1 race to and from every destination it would do the same...I currently have a great dyno tune that is keeping me going and my ratios/temps have remained great even with adding more parts, i could benefit from doing a retune, sure, but scheduling it (4 hours seems to be the average locally) is hard with work, kids and all the life that goes with it...i agree on the blackbox theory, also adds concern of the potential for someone to hack my car, bad tune is loaded and engine is trashed. i think the AI could spot pranksters loading fake data as there would be no feedback from those vehicles to the system, much less data would be different streaming it, simple OBD2 code readers can read the VIN and thereby allow the system to verify vehicle data/engine type and other sensor data...monthly subscription costs, sure, if it works and does the job correctly...
I wouldn't necessarily work like that... A well implemented setup would constantly correct for targets... NOT your driving style. It also wouldn't spot "pranksters" any more than a current tuning implementation does.

What I would worry about is how it would react to actual issues with the car. If you were to spring a leak, or have something veer off, how much correction would it apply before it gives up? As for selecting drop-downs for parts installed on a car, some tuning software already does this... quite a few actually, just not Subaru.

Ideally, this seem more like short trims having a larger affect on long trims to the point that it "re-writes" the tune. I could see this being good if everything is right with the car, but potentially just a catastrophic if not. Additionally, when it comes to making the best, efficient power, how would this be implemented? How it is going to determine best AFR, timing, cam timing, etc? With ambient temperature, pressure, etc changes, will you still involve a proactive approach as well as it trying to adjust after the these events are seen?

It's definitely interesting, however, with Subaru's being so finicky and easy to blow, what would be in place for when that does happen.
 

·
Premium Member
Joined
·
5,331 Posts
i think the AI could spot pranksters loading fake data as there would be no feedback from those vehicles to the system, much less data would be different streaming it, simple OBD2 code readers can read the VIN and thereby allow the system to verify vehicle data/engine type and other sensor data
VIN ect would help - with base vehicle. not mods, and certainly not malicious "users" they wouldn't even need a car! All the need is some programming and any computing base, a PC or even a raspberry pi :)

I look at this more as an auto logging AP - so I don't feel guilty of how long I go between log and review cycles. Ya know, Cobb is reading this:)

But back to the tune: How bout it works and a windows update, except you get to review the changes an load them after approval?
 

·
Registered
Joined
·
2,256 Posts
Discussion Starter #9 (Edited)
Sorry guys, my inbox has been full, I just cleared some of it out. ML works best at scale. My initial thoughts are, for this to work effectively, it would have to be partnered with pro tuners instead of directly competing or largely open-source with some premium services. With enough sets in the model, outliers or setups that don't respond as expected to inputs could be pretty easily identified to prompt the user to update the mods/info or troubleshoot. I need to think about how a lot of this could be done open-source.

I think the real benefit and disruptive aspect of this would be the ability to modify the cars with individual components rather than in batches because of the additional time and money costs of tuning. I'm not sure that's of value to most though as the vast majority only do simple bolt-ons that canned tunes cover... it becomes a question of causality.

It's really an ideal application of AI/ML it's just a question of value and whether or not it could get to the scale needed to work properly.

I wonder if there is some way to incentivize tuners to share logs and maps with some sort of crowd-sourcing payout.
 

·
Super Moderator
Joined
·
8,410 Posts
Sorry guys, my inbox has been full, I just cleared some of it out. ML works best at scale. My initial thoughts are, for this to work effectively, it would have to be partnered with pro tuners instead of directly competing or largely open-source with some premium services. With enough sets in the model, outliers or setups that don't respond as expected to inputs could be pretty easily identified to prompt the user to update the mods/info or troubleshoot. I need to think about how a lot of this could be done open-source.

I think the real benefit and disruptive aspect of this would be the ability to modify the cars with individual components rather than in batches because of the additional time and money costs of tuning. I'm not sure that's of value to most though as the vast majority only do simple bolt-ons that canned tunes cover... it becomes a question of causality.

It's really an ideal application of AI/ML it's just a question of value and whether or not it could get to the scale needed to work properly.

I wonder if there is some way to incentivize tuners to share logs and maps with some sort of crowd-sourcing payout.
Let me know if you want some help from the the tuning side. It is probably an idea that can work, but will need to span across the tuning community.
 

·
Premium Member
Joined
·
5,331 Posts
My initial thoughts are, for this to work effectively, it would have to be partnered with pro tuners instead of directly competing or largely open-source with some premium services.
!!!!!!! Awesome idea! they get to charge and do less work! Only problem is you have to share the funds and it will cost time, or money, or both to develop.

I think the real benefit and disruptive aspect of this would be the ability to modify the cars with individual components rather than in batches because of the additional time and money costs of tuning.
Agreed!

I'm not sure that's of value to most though as the vast majority only do simple bolt-ons that canned tunes cover... it becomes a question of causality.
Sure but its the "builders and owners of built cars that that would definitely pay. Like Cobb and other tuning platforms, it will need to cross platform to be profitable. A least the following cars would be less expensive to develop.

It's really an ideal application of AI/ML it's just a question of value and whether or not it could get to the scale needed to work properly.
* Why ideal for AI?
* Covered

isn't this basically already done via cobb?
No, at least not my understanding.

Let me know if you want some help from the the tuning side. It is probably an idea that can work, but will need to span across the tuning community.
In for help if there is something I can do. Might be able to help with hardware.
 

·
Registered
Joined
·
2,256 Posts
Discussion Starter #14
!!!!!!! Awesome idea! they get to charge and do less work! Only problem is you have to share the funds. It will cost time, or money, or both to develop.



Agreed!



Sure but its the "builders of owners of built cars that that would definitely pay. Like cOCobb and other tuning platfroms, it will need to cross platform to be profitable. A least the following cars would be less expensive to develop.



* Why ideal for AI?
* Covered



No, at least not my understanding.



In for help if there is something I can do. Might be able to help with hardware.
Honestly, I'm not interested in building some big enterprise like Cobb. I'm more interested in building something disruptive and beneficial to the tuning world that maybe I can make enough off of to maintain my lifestyle while working on it full-time.

@yamahaSHO, do you know if it's commonplace for tuners to keep logs and maps for the cars they tune in any kind of database?

Thanks for the interest and support, guys. This is just a concept right now. This is a space that has a lot of very technical people, but doesn't seem to have applied many of the modern tools available and used in other spaces. I think there's definitely something here, just a question of how viable a solution would be and what form it would take. Even if it ends up being something completely open-source, that would be cool.

PS sorry my inbox filled up again
 

·
Super Moderator
Joined
·
8,410 Posts
off of to maintain my lifestyle while working on it full-time.

@yamahaSHO, do you know if it's commonplace for tuners to keep logs and maps for the cars they tune in any kind of database?
I keep logs of just about everything. The big part is setting up the base map first, which will include having all the proper input scaling and such. These days, places like injector manufacturer's already give you data.

After that, it's build specific on how much air is flowing through and where it is most efficient.
 

·
Registered
Joined
·
3,132 Posts
before you get any value out of this, you will need a significant amount of learning data. How would you incentivize people to provide this and keep their patience while the AI learns.

I think this would be fairly easy to build. As i mentioned in the PM, i've already mostly built the data collection and processing infrastructure you would need to do this.. Just add code modules.

however, i think there is a bigger opportunity in this and AI tuning is just one use case for data analytics.
 

·
Registered
Joined
·
2,256 Posts
Discussion Starter #18
before you get any value out of this, you will need a significant amount of learning data. How would you incentivize people to provide this and keep their patience while the AI learns.

I think this would be fairly easy to build. As i mentioned in the PM, i've already mostly built the data collection and processing infrastructure you would need to do this.. Just add code modules.

however, i think there is a bigger opportunity in this and AI tuning is just one use case for data analytics.
That's why I'm interested in the idea of incentivizing tuners to contribute base maps and log data. If we could accurately attribute value to data and map contributions then offer commensurate incentives through some sort of distributed payout mechanism, training data could scale really fast.

I'll email you about what we've discussed.

Again apologies for the PM slowness, everyone. I don't have a lot of free time during the week right now.
 
1 - 18 of 18 Posts
Top