New Tuning Method?

[Silvo 1981]

I've been using this as method for tuning my cars for a while now but when I posted this on FM.net I didn't get any responses. I think it's a useful technique but I'm not sure if others have used it or if they think it's any good.

The hired driver technique...

Before you start the race you get the option to hire a driver driven by the computer instead of racing the race yourself. By using the hired driver you get a couple of things; very consistent lap times (once you have hit clean air in front) and time to get on with the things you've been putting off doing at home while still developing your garage. You don't have to race the whole race but you need to get in front all the cars and have at least one lap without other cars holding you up. I generally do this by decreasing the difficulty settings to easy and having a higher PI than the competition.

Take note of the fastest lap time achieved, make an alteration to the build/tune and re-test at the same race again. If the lap time improves then it was a positive change and should be kept, if it doesn't then you can revert back and try something else.

I mainly use this technique to get the build right as I still want to tune myself. However, sometimes it's difficult to tell if an improved lap time is due to the small differences in the build (ie lighter rims for decrease in power whilst keeping the same PI) or familiarisation of the track and/or inconsistencies in my driving. By letting the computer drive the laps for you you get a good consistent lap time (pretty much to the hundredths) and you don't have to sit there and do the hard work freeing you up to do the things you should be doing!

[fndrbndr]

Great idea, as long as I'm understanding it correctly.  You're saying to use this for build only, not for the tune?  'Cause I can't imagine driving like some of the computer drivers do, and wouldn't want a car tuned to their methods.  Do the results mesh fairly well with results from driving the same events yourself with multiple tunes?  I ask because the computer tends to do better at the extremes than I do, braking at the last possible second and putting down exactly as much power as the tires will handle, while I tend to pick better lines than the computer.  My guess is that really twitchy, high-powered cars might be MUCH faster with computer drivers, while human drivers would struggle just to keep them on the track.  Of course, I could also easily be wrong on this, just trying to think about how the differences might have a systematic effect on your findings.

[Spiny Anteater]

Got to agree with fndr on this one - while I can see the benefits for the build I think there is just too much variability between humans for the tuning, let alone between humans and the CPU. You just have to look at us here. Some prefer a much more understeery setup than others and if 8 of us tune the same car for the same track in the same class, you can pretty much guarantee 8 different tunes, sometimes wildly so. Thing is, we usually end up with similar builds though - I guess that's where the CPU aspect can come in handy.

[Silvo 1981]

I've only really used it for building and haven't done any tuning with it for the type of reasons you are saying. I probably need to do some testing to see if I can better the build I come up with using the hire driver method but the results I had building this way were definitely some of my best in FM2 and I'm sure I will use it in with good results in 3 as well. I just thought I'd share the idea to see if anyone else had tried it or used it because I couldn't find anything on it.

The computer is generally slower and you cannot compare it with the times you do yourself but you can compare it against itself and therefore you can judge changes to the build quite easily and without much effort.

[Blooze]

That would have been a good method for my "Princess and the Pea" project.  Wish I woulda thought/heard about it back then.

$

[donpost2]

I'd like to revive this topic if possible 

First of all let me compliment you on your forum, very useful. I discovered this place after reading a thread by Fit on the official forums, and I've learned a lot from here already.

I always like to take a scientific approach to things, so this thread has clearly piqued my interest. Last night I tested out this idea for it's feasability, and I came up with the following pro's and con's

Pros

• In my testing, I let the AI loose and recorded the fastest lap time. I repeated this three times. The spread of times was never more than 0.15 of a second, so certainly accurate and consistent enough for this task
• I tested a few different builds using this method (In D-class), and the difference in times between the 3 builds were in the order of 0.3 of a second, so the results do vary enough for this method to identify the best build. However, I'm not sure how reliable it would be if the builds were only slightly different.

Cons

• It takes bloody ages! Especially if you are repeating the test of each build 3 times
• Unless you know a way to get the AI driver to drive on tracks other than the ones listed in the event's list we're going to be pretty limited as to which track/class combinations we can use this method on

Thoughts? The biggest problem in my eyes is con number 2.

[feuerdog]

I've used this kinda method before for build evaluation.

I'd build 3 cars based on the same class/chassis/PI/settings/etc.
I'd tweak one for grip, usually adding aero and a tire upgrade.
I'd tweak another for power, just the opposite of the grip car.
The third car would be left alone, as per my typical build theories, as a baseline.

I'd set the drivatar to run three laps with each car, add the lap times for each, and see which car had what strengths for a given track.

Most of the time the differences were small, but all of the time the effort was large.
The biggest problem is that the AI doesn't take advantage of the entire track, it simply runs a generic line.
We, as actual drivers, can anticipate corner combos and take more advantage of car and track specifics to negate most of the drawback differences.

Does it work on a comparitive level? Yes.
Do I personally use this method,...not anymore.

I'd rather just run a different car around the same track myself, and then compare builds and times in person.

[donpost2]

I'd rather just run a different car around the same track myself, and then compare builds and times in person.

That would be the easiest way, but the problem I personally have with that method is that I am not consistent enough in my laptimes for the comparison to be fair.

Maybe since, broadly speaking, you can classify tracks into one of three types, as long as one of each type is available for the AI to race a method could be devised....

*revisits drawing board*

[Spiny Anteater]

...
That would be the easiest way, but the problem I personally have with that method is that I am not consistent enough in my laptimes for the comparison to be fair.

Maybe since, broadly speaking, you can classify tracks into one of three types, as long as one of each type is available for the AI to race a method could be devised....

*revisits drawing board*

I hate to sound rude, but that is probably the first place to work on. It's good that you're working out the principles of tuning, and there is no doubt that they will help you as you progress in the game. And if your main interest is in tuning then by all means keep going as you are - after all the only wrong way to play this (and other) games is in a way which you don't enjoy.

But, if you're trying to find more speed, you will find more of it in tightening the nut behind the wheel than you will in the thousands of other nuts and bolts which make up your car (if it's still around, Big Moo's excellent thread on fm.net a couple of years ago explained this brilliantly). You don't need to be consistent to within a tenth of a second to see an improvement in a tune - if you can run 5 laps within 2 or 3% of each other that will give you a decent indication of if a tune has improved or not. And if you're doing the test runs, you're sharpening your skills and getting a tune which fits you which will make you and your car faster than just watching the AI driving around for you.

(Sorry, this isn't meant as a negative post, because you are doing the right thing in thinking about what's happening and reading around the subject. It's just I think if you drive the track rather than the AI you'll make progress faster

[Fit4aking]

Good to see you over here donpost! Welcome to the site and I'm glad it has some information you can use.

I'll have to agree with both feuer and Spiny on this one.  The mothod makes absolute sense to eliminate yourself as a variable and it gives you laps of telemetry to watch and take in but if you take the car off of the CPU's line you may find issues that you never would see from "his" driving style.  If you aren't pushing the car to your limits, regardless of consistency, you'll never know what is really plaguing the setup.  I know for a fact the drivatard brakes early and that isn't my style so tuning for "him" isn't ideal for me. 

Running the car yourself in a private lobby filled with AI will allow you to save the replay, run on any track, for as many laps, and in whatever cars you want.  Along the way you'll be getting loads of seat time, adding to your bank account, and getting familar with the cars.  More positives than your short list of negatives.  Tuning isn't quick but once  you grasp it you can knock lots of time off of your setups.  I surely have.

Oh, and I too have gone back to the drawing boards over time and made leaps because of small steps backwards.  Practice!

[donpost2]

Thanks for the response - so quickly too!

After reading what you have to say, and doing some thinking, and trying the idea out practically - I've come to the conclusion that this is definitely a no go.

When it comes to my driving, yeah its not uber consistent, so I guess improving that will save me more time that tweaking hundreths of a second with the build.

Thinking about it, it's not too difficult to figure out what direction your build should take for the track. I only describe my builds in terms of three variables anyway: Handling, Power, and Power-to-Weight ratio. It's not hard to figure out which variables to focus on in the tune considering the track: Eg. Tsukuba - You might think that more handling is better, but all bar one of the corners are real slow, so better handling isn't going to give you much of an advantage. Instead, you have three medium length straights - so power-to-weight is the way to go, with perhaps a touch of handling just to ensure you get a decent speed through the final sweeping turn. Sound like a correct analysis?

Also, after trying the idea out in practice I found there were a few problems. As I said, finding the correct event is a headache, and also when you are racing the other AI races can interfere with your AI driver if they have a similar speed build, and it makes collecting laptime data difficult due to things like drafting.

Thanks again for your input guys  I'm sure I'll be posting a bit here and there, mostly with questions!

[Silvo 1981]

I tend to only use this method if the xbox is running in the background and i'm supposed to be doing something else. It's certainly not the only method I would use to build but it has been a method I have had decent results with and thought I'd share it to see if others had thought of it or were using it with any sucess.

[Drift2XL]

I think that the fact that we are talking about builds here. And from my experience, is 75% of the tune. If your wondering if this build is faster than that one it is helpful.

But remember what the drivatars weakness are. Early braking, mid corner brake checks, less than optimal corner speeds. Then there is the fact that they really can't handle pure missile builds too well. Which will skew the results.

My basic philo is build pure grip. If you drive it a lot, you'll anticipate its weaknesses and work around them. Then start to take away grip and add power. On certain tracks, if you're not feeling it is a little hairy, you need more power.

But then this is for a car your going to drive a ton.

[Spiny Anteater]

...
Thinking about it, it's not too difficult to figure out what direction your build should take for the track. I only describe my builds in terms of three variables anyway: Handling, Power, and Power-to-Weight ratio. It's not hard to figure out which variables to focus on in the tune considering the track: Eg. Tsukuba - You might think that more handling is better, but all bar one of the corners are real slow, so better handling isn't going to give you much of an advantage. Instead, you have three medium length straights - so power-to-weight is the way to go, with perhaps a touch of handling just to ensure you get a decent speed through the final sweeping turn. Sound like a correct analysis?

...

Sounds good in theory. But...

A couple of years ago in FM2 we had a site challenge running NSX's around Tsukuba. Now, a good number (probably the majority) of the guys here at the time tried that theory first of all. But, it turned out that the pure grip builds were the fastest. What tends to be forgotten is that the slow corners actually place a premium on MECHANICAL grip (i.e. tyres) - on a track like Tsukuba with only very short straights a high power-to-weight car will spend most of the straight trying to catch up the speed lost in the corner compared to a grippier one.

What I'm trying to say is, if you're hotlapping, a high-grip car will be your fastest option at more tracks that you would expect just because we naturally tend to forget how much time we're losing in the corners and subconsciously focus on that higher top speed. I should also mention though that if you're planning on going racing, the corollary of that extra grip is usually a low top speed and that will leave you a sitting duck on the easiest overtaking locations - the straights.

But, don't let me discourage you. You're right that you can usually anticipate the build required for a track. Just be aware that we tend to underestimate the impact of extra grip. And also make sure you decide whether you want a car for hotlapping or racing.

[Choccy]

...
Thinking about it, it's not too difficult to figure out what direction your build should take for the track. I only describe my builds in terms of three variables anyway: Handling, Power, and Power-to-Weight ratio. It's not hard to figure out which variables to focus on in the tune considering the track: Eg. Tsukuba - You might think that more handling is better, but all bar one of the corners are real slow, so better handling isn't going to give you much of an advantage. Instead, you have three medium length straights - so power-to-weight is the way to go, with perhaps a touch of handling just to ensure you get a decent speed through the final sweeping turn. Sound like a correct analysis?

...

Sounds good in theory. But...

A couple of years ago in FM2 we had a site challenge running NSX's around Tsukuba. Now, a good number (probably the majority) of the guys here at the time tried that theory first of all. But, it turned out that the pure grip builds were the fastest. What tends to be forgotten is that the slow corners actually place a premium on MECHANICAL grip (i.e. tyres) - on a track like Tsukuba with only very short straights a high power-to-weight car will spend most of the straight trying to catch up the speed lost in the corner compared to a grippier one.

What I'm trying to say is, if you're hotlapping, a high-grip car will be your fastest option at more tracks that you would expect just because we naturally tend to forget how much time we're losing in the corners and subconsciously focus on that higher top speed. I should also mention though that if you're planning on going racing, the corollary of that extra grip is usually a low top speed and that will leave you a sitting duck on the easiest overtaking locations - the straights.

But, don't let me discourage you. You're right that you can usually anticipate the build required for a track. Just be aware that we tend to underestimate the impact of extra grip. And also make sure you decide whether you want a car for hotlapping or racing.

Just to expand on Spiny's point. Cornering is a combination of steps including setting up for the corner, braking, entering the turn, navigating the turn and exiting under acceleration. A car with more grip helps in all aspects of cornering, but one of the under-appreciated techniques of road racing is the ability to get on the power earlier than your competitor.

Nailing your apex and allowing the car to track out while squeezing the accelerator without experiencing slip pays dividends because you are on the power earlier. Meaning you're getting your car to it's highest possible  speed along the straight more quickly than a competitor.

A car with a higher top-end and less grip can be wrestled with to earn a win, but you'd be surprised how well a less powerful car can do if the driver capitalizes on what's available.