BRC 1966 - Gentleman Brobots Club [RACE 6 P&Q]

[mer_at]

Sooo basically, what we need is a racing series with 100 races, and to replay that series 10 times to see how the data compares.

Almost right.
Problem is - we have only 8 races:
- built a car for slow tracks
- hey look, slow tracks incoming, oh the joy ...
- get rekt on 2 races thanks to <random> crashes
- aaaaand season is gone ...

Makes me wonder why i did invest all my time in that car.

If you have only 8 races, drop the crash rate! I want to see how my CAR DESIGN competes against other peoples CAR DESIGNS.
If i want to see HOW LUCKY I AM, i just go to the casino and play some roulette .....

And Norman got rekt even worse throughout BRCs ...

[AirJordan]

OK I'm really no big fish (yet?) around here but since we are sharing opinions...

Situation is: my drive/sport is almost 1, driveability good, safety average, reliability lowish, weight low and being slow I'm constantly crowded. I finished only one race so far.

About the statistics, killrob is absolutely right, as far as "science" can be absolutely right. You can not judge random values based on 4 events. I mean yes you can do whatever you want but your results are meaningless. And if you do not have a plan to invent something new it's a not up to debate, sorry.

As far as being constantly out, I would say that it seem a bit too harsh for roughly 30% of the field do not finish. Maybe it should be more dependable on actual specs or simply a bit easier. Especially because season is so short.
Next suggestion for the future would be possible revisions mid-season, which is quite realistic. You could limit the to only few parts or so...

Nevertheless I think BRC is awesome and a appreciate that some of you guys (devs & info gurus) are spending a lot of your time on this. I really wish that there would be more BRCTool challenges where you could see the simulation in real time.

[Leonardo9613]

damage in crashes: dmg ~ random / durability * opponent_mass / own_mass

damage in errors: dmg ~ random / durability
error chance ~ TrackSportiness / TrackTameness * (rain + constant) --> more errors in rain.

I wonder how big is the random value on this case, which could explain why the randomness is dominating over the constant parameters.

For the future I want to revise the error/damage/crash calculations. In a first draft yesterday I revised the damage taken in crashes: First an overall crash severity/energy is calculated (randomly). This energy is then distributed to the victims: 33% of the energy to each car + divide the remaining 33% between the two cars. Then we know how much energy the car has to absorb and safety or reliability can come into play: total taken damage is then energy * (100-safety)/100.

Sincerely, rather than using a random split, I think you should use some conservation of momentum and then consider the rate with which the momentum was changed as the damage value. For example, car 1 is doing 150 kph and car 2 is doing 100, considering a perfectly elastic crash, let's say that car one would have sped up to 120 kph, where as car two has slowed down to 125 kph. Car one had a 20% change in momentum, where as car two had a 16.6% decrease. You could use something based on that to estimate the damage more accurately, at least with some degree of physics involved, so light scrapes would just be that, and big shunts would have big results.

It's fine, my car is doing well

My concern isn't just because my car, norman's, merat's and cobaltgirl's cars have crashed out time and time again, it is because we deliberately made our cars in a way that we understood would make them less vulnerable to such events. If not being in the mid-field is what it takes to not crash, then everyone should have made 0 safety cars with low reliability, which are lighter and cheaper, thus allowing us money to spend elsewhere. My car, for example, could be 2s a lap faster if I didn't try to play it safe when it comes to tyre wear, safety and reliability. To put it in one sentence, it feels as if I'm being penalised for being cautious, cautious trying to avoid exactly what is happening to me right now.

[RobtheFiend]

If the number of driver errors in the practice on Thruxton is an indicator, maybe half the cars will crash themself out.
Is it just me, or where there many more driver errors than the previous 4 practice sessions?

[Puffster]

I just keep being impressed by the performance of Pyrlixs car, truly remarkable how well it performs allround on so many different tracks!
I do think there is something to be said about having a teal coloured car, soothes you out of the worst crashes

[RobtheFiend]

I just remembered, mine had 5.8kg of fuel, Vikus had 9.5kg. I wonder how much he would use if his car would survive a full race.

[Absurdist]

I like Der Bayers proposed change.

4th One up from last race.

Now I've gotta choose between another one stop or a two stopper. Packbat! Heeeelpp meeeeeee XD
I didn't like the one stop, but It got me my best result so far. Might try it again, hope I jump everyone at the start and hold them up again. But if it rains that'll be a disaster.

Hmm...

Grats Puff on the 2nd, nice.

[Packbat]

...this is turning into a long post, so a quick thing before everything else: Is the practice/qualifying data from Thruxton in the .zip in the OP? When I downloaded it yesterday (about nineteen hours before this post), it had the Oulton Park race results, but none of the Thruxton .csv files.

That aside: I've been avoiding the thread for the past couple days because I wanted to come in after having proven that something on the order of 330 vehicle-hours of data is enough to show that there's a real issue with the damage system ... but while I'm pretty sure that I still could, Der Bayer has probably saved me the other three hours of data mining that it would take. So thank you very much.

First, regarding that aforementioned issue with the damage system: I don't know how your random-number generator works (is it a normal distribution? What's the mean and standard deviation?), but I feel like a reference point that people would really like to have (and, worse, thought they did have, approximately) is the five-percent point:

• If you have X durability, then you will have a 5% chance of being eliminated by any given driver error.
• If you have Y durability, then you will have a 5% chance of being eliminated in an crash with a car of equal weight.
• If you have Z engine reliability, then you will have a 5% chance of being eliminated due to engine failure in a race of this length.

The sense I get is that many contestants (myself included, to be honest) thought the most durable cars would have about a 1% chance of being eliminated per crash and the least 10% or 20% or 50% or more ... whereas what actually happened is that the most durable cars had around a 20% chance versus the least around 30%. In other words, the problem is that we thought we would be able see the difference in crashworthiness between the worst and best cars in just four races, and were putting a lot of work into judging the compromise between durability and speed on that basis.

Second, I absolutely love your planned revisions - especially the driver-pressure system. I just want to go ahead and throw a couple quick suggestions into the pot:

• More stress means more errors, but more stress also means more speed. The errors will probably be more frequent and possibly more severe, but between the errors the driver should be driving a little faster than they would alone.
• While the pursuit is less stressful than the fight, it is still stressful. I've seen enough sim-racing to see that drivers make more errors when they are trying to catch someone (or pull away from someone) than when they're just lapping out by themselves. Perhaps there should be a small amount of pressure applied to both cars as they approach within ten seconds of each other, and then a larger amount of additional pressure to the lead car as the distance closes to within one second.
• You mentioned that drivers feel pressure by a car close behind. I would revise that to say: drivers feel pressure by a car close behind at a similar speed. If the car coming up behind is much faster than you, then there's no point in stressing out because they'll pass you anyway. (Likewise for the trailing car, naturally - a slower car up ahead isn't competition, it's traffic.) I'm thinking "similar speed" would probably be something like "less than two seconds a lap difference on a six-kilometer track", but that's a judgment call - and it probably shouldn't be a binary on-off switch anyway.
• For crashes: the random amount energy should probably be scaled by speed. I imagine you're already planning to do that, though.

[Absurdist]

Trust in thy Der Bayer. He will lead us to paradise!

Can there be waffles?

[Der Bayer]

Is the practice/qualifying data from Thruxton in the .zip in the OP?

You're right, the data was missing. I added it now.

First, regarding that aforementioned issue with the damage system: I don't know how your random-number generator works (is it a normal distribution? What's the mean and standard deviation?), but I feel like a reference point that people would really like to have (and, worse, thought they did have, approximately) is the five-percent point:

• If you have X durability, then you will have a 5% chance of being eliminated by any given driver error.
• If you have Y durability, then you will have a 5% chance of being eliminated in an crash with a car of equal weight.
• If you have Z engine reliability, then you will have a 5% chance of being eliminated due to engine failure in a race of this length.

I use mostly an equally distributed random number generator (between 0 and 1). I do agree that giving a general feedback on the chance of errors/damage/engine failures will give the participants a better feeling for what is happening and for when they just have been unlucky. I will add this in the next challenge annoucement. And if I don't, please remind me to do it

• More stress means more errors, but more stress also means more speed. The errors will probably be more frequent and possibly more severe, but between the errors the driver should be driving a little faster than they would alone.

Hm, I don't like drivers going faster than their optimal speed (that's basically what they are doing now) just because they are under pressure. I think that they even should lose time because they are tyring to cover position. The system I have in mind works with opponents being very close (< 30 m), so that's when they really should watch out.

• While the pursuit is less stressful than the fight, it is still stressful. I've seen enough sim-racing to see that drivers make more errors when they are trying to catch someone (or pull away from someone) than when they're just lapping out by themselves. Perhaps there should be a small amount of pressure applied to both cars as they approach within ten seconds of each other, and then a larger amount of additional pressure to the lead car as the distance closes to within one second.

The first pressure system I want to introduce is the system where a leading driver is under pressure by a following driver who does not have the straight line speed to pass but is faster in general. I see your points but let's first get this neccessary mechanic in and working and then discuss further improvements.

• You mentioned that drivers feel pressure by a car close behind. I would revise that to say: drivers feel pressure by a car close behind at a similar speed. If the car coming up behind is much faster than you, then there's no point in stressing out because they'll pass you anyway. (Likewise for the trailing car, naturally - a slower car up ahead isn't competition, it's traffic.) I'm thinking "similar speed" would probably be something like "less than two seconds a lap difference on a six-kilometer track", but that's a judgment call - and it probably shouldn't be a binary on-off switch anyway.

You're right. I prototyped a system where it takes some time until the driver is stressed a lot. So if a car is approaching very fast and overtaking soon, the driver won't be stressed much (and will relax afterwards anyway because the opponent is gone). In reality a driver can encounter the attacks for a shorter or longer time until he makes an error. So over the period of a few minutes of heavy pressure by a close opponent his stress level rises. This is what I try to implement.

• For crashes: the random amount energy should probably be scaled by speed. I imagine you're already planning to do that, though.

This is a little bit more difficult because the speed differences are very low in general. Because the drivers don't make real errors (braking too late etc.) I calculate crashes on a very rare case scale when cars are very close to each other and in corners. The crash energy is an arbitrary value then. But the first revision of the system already gives a lot more small collisions without much damage and only a few big ones. The ones being hit will still whine but that's racing and I won't dismiss the random elements because then we could base everything on comparing lap.

To sum it all up: Thank you for the very detailed and constructive criticism! Much appreciated.

[Packbat]

You're welcome - and thanks for the detailed response!

Is the practice/qualifying data from Thruxton in the .zip in the OP?

You're right, the data was missing. I added it now.

Sweet! Tyre wear and fuel consumption calculations atta-holy crap there's a lot of tyre wear on this track. Wow. Attached.

5_Thruxton_P-90 min.zip

(23.99 KiB) Downloaded 296 times

I use mostly an equally distributed random number generator (between 0 and 1). I do agree that giving a general feedback on the chance of errors/damage/engine failures will give the participants a better feeling for what is happening and for when they just have been unlucky. I will add this in the next challenge annoucement. And if I don't, please remind me to do it

So ... durability 100 = 0% chance of being wrecked out in a single collision, durability 54.6 means 45.4% chance of being wrecked out in a single collision, and durability 12.4 means 87.6% chance of being wrecked out in a single collision (more or less). Yeah, that's a much weaker dependence on reliability than I think we were expecting.

• For crashes: the random amount energy should probably be scaled by speed. I imagine you're already planning to do that, though.

This is a little bit more difficult because the speed differences are very low in general.

I was actually thinking absolute speed, not relative speed - but then again, a lot of race-ending incidents happen in corners at low speed, so never mind, using a fixed energy scale is probably good.

But the first revision of the system already gives a lot more small collisions without much damage and only a few big ones.

Awesome! I look forward to seeing it in action.

[TheBobWiley]

So will we be able to do something to increase the "stress time" for our drivers? That luxury 8 track and Luxury interior could use some love

[Packbat]

Sweet! Tyre wear and fuel consumption calculations atta-holy crap there's a lot of tyre wear on this track. Wow. Attached.

...I gotta do it. I'm going for a two-stop strategy. If I tried to do a one-stopper, I'd be pushing close to 30% tyre wear, and that would cost me just too much time.

Wow. Huge advantage to the low-tyre-wear vehicles at this track.

[rcracer11m]

So will we be able to do something to increase the "stress time" for our drivers? That luxury 8 track and Luxury interior could use some love

Something I had thought of was having a price for your driver. There would be multiple different people to choose each with their own strengths and weaknesses and would make you think whether you would want a faster more expensive car with a bad driver, or a slower cheaper car with a better driver. Probably wouldn't happen though as it would most likely be very difficult to implicate

[Killrob]

So will we be able to do something to increase the "stress time" for our drivers? That luxury 8 track and Luxury interior could use some love

Something I had thought of was having a price for your driver. There would be multiple different people to choose each with their own strengths and weaknesses and would make you think whether you would want a faster more expensive car with a bad driver, or a slower cheaper car with a better driver. Probably wouldn't happen though as it would most likely be very difficult to implicate

That's a nice idea as it offers a new compromise and such, but would also be really hard to balance.