do engine/tire/chassis choices affect AI performance?

[samsepi0l]

Let's say we have 2 drivers with the same values in DRIVERS2.TXT

take - for example, Michael Andretti, Alex Zanardi and Jimmy Vasser below:

VERSN 1
NUMDR 40
NUMCH 3
NUMEN 3
CHASS _ _ _
ENGIN _ _ _
NUMT 2
TIRES Goodyear Firestone
DINFO 0 2 0 0 375 562 375 562 387 562 375 562 Pace_Car Pacer Denver,_CO USA USA INDYCAR
DINFO 0 2 0 1 500 562 500 562 100 150 998 999 Peter_White 5/10/1969 Reynard_96I/Honda_961_HRH Austria AUS Blanche_Racing
DINFO 12 2 2 1 445 550 445 550 395 570 750 850 Jimmy_Vasser 11/20/1965 Reynard_96I/Honda_961_HRH USA USA Target_Chip_Ganassi_Racing
DINFO 32 0 2 1 375 530 375 530 400 575 525 610 Adrian_Fernandez 4/20/1965 Lola_T9600/Honda_961_HRH Mexico MEX Tasman_Motorsports_Group
DINFO 8 2 2 0 375 535 375 535 395 570 750 850 Gil_deFerran 11/11/1967 Reynard_96I/Honda_961_HRH Brazil BRA Hall_Racing
DINFO 20 0 0 1 320 515 320 515 400 575 500 700 Scott_Pruett 3/24/1960 Lola_T9600/Ford_Cosworth_XB USA USA Patrick_Racing
DINFO 6 0 0 0 445 550 445 550 395 570 940 999 Michael_Andretti 10/5/1962 Lola_T9600/Ford_Cosworth_XB USA USA Newman/Haas_Racing
DINFO 17 2 0 0 315 505 315 505 400 575 500 650 Mauricio_Gugelmin 4/20/1963 Reynard_96I/Ford_Cosworth_XB Brazil BRA PacWest_Racing_Group
DINFO 4 2 2 1 445 550 445 550 395 570 900 999 Alex_Zanardi 10/23/1966 Reynard_96I/Honda_961_HRH Italy ITA Target_Chip_Ganassi_Racing

(this is just a clip from my drivers2.txt, the rest is not needed)

Notice how the values for performance on all 3 drivers are the same (aggression value is ignored) 445 550 445 550 395 570

I would think, that all drivers would usually perform about the same. Over enough races I would assume that all 3 of these drivers would be equally successful. For some reason in my season I am running, that is not the case. Jimmy and Alex have been beating Michael pretty much consistently.

I'm wondering if the difference is due to the engine/chassis and tires? I do not know if certain engines/tires/chassis (I'm speaking FOR THE AI ONLY) have a performance advantage- or does the game just use the numbers in the drivers file? I was contemplating giving all drivers in the file the same engine and chassis so that it would ignore this if it were indeed happening. Obviously, I don't want to give everyone the same tires as you can physically see a difference in the cars appearance based on tires choice. I use a descriptive field in the drivers file to make the engine/chassis combo read out how I want anyways, so when "viewing opponents in carset" it can read any engine/chassis combo I want regardless. I also make the selections in the file an underscore charecter "_" so the engine or chassis does not show up in the visual when the car is spinning viewing opponents (you have to set the file to read-only to do this).

Anyone have thoughts about this? In the 96 season Andretti/Vasser and Zanardi were really the key players so I wanted them to all have equal performance. This is basically true in the 96 season during real life. I'm just trying to find out why I've run 13 races at full 100% so far and Michael has only won once compared to 5 with Jimmy and 3 with Alex.

[CowtownBob]

(Hopefully this will make sense - bear in mind that it's the morning after)

Yes, engine/chassis/tire values make a difference. They act as modifiers to the equations that calculate the AI's movements (actually they effect your car too).

But why are you ignoring Aggression when it's part of the answer you're looking for? Aggression works as a negative value - the lower the numbers, the more aggressive the driver. So in your example, Vasser is the most aggressive with a range of 750-850 and Andretti the least at 940-999. Your race results seem to reflect this.

For each race, each driver will be assigned a value from the range specified in drivers2.txt. Vasser's aggression value will always be substantially lower since the top of his range (850) is lower than the lowest value for either Zanardi (900) or Andretti (940). On an individual race basis, the difference could be as much as Jimmy running at 750 while Alex and Mikey are at 999.

Keep in mind that the carset you build will not be a perfect recreation of historic results, but rather an approximation of the possibilities/probabilities of what could have happened. Your top three drivers have won 70% of the races you've run, which seems pretty good.

[samsepi0l]

Yes, engine/chassis/tire values make a difference. They act as modifiers to the equations that calculate the AI's movements (actually they effect your car too).

ok, what I will do is give everyone a Reynard/Honda - because I want the performance to be based solely on the numbers (since we don't know the mathematical effect the modifiers). It will still be listed as text in the carsets as the correct cars for the specific year. As far as tires go- I still need to tweak the tire values in the drivers text right? I thought the firestones have more grip but the goodyears lasted longer (according to what I've read). Can't I just copy the values from tire2 and put them in the first tire spot?

But why are you ignoring Aggression when it's part of the answer you're looking for? Aggression works as a negative value - the lower the numbers, the more aggressive the driver. So in your example, Vasser is the most aggressive with a range of 750-850 and Andretti the least at 940-999. Your race results seem to reflect this.

For each race, each driver will be assigned a value from the range specified in drivers2.txt. Vasser's aggression value will always be substantially lower since the top of his range (850) is lower than the lowest value for either Zanardi (900) or Andretti (940). On an individual race basis, the difference could be as much as Jimmy running at 750 while Alex and Mikey are at 999.

I always thought that the aggression value was the likelihood that they would try a pass? I am going back 20+ years, but I remember back in the day Paul Tracy and Robby Gordon had high values in the game by default- which in real life those 2 were always getting into trouble (especially Tracy). Zanardi is/was super aggressive (think of the Laguna seca "the pass" move). The thing that gets me is I could have sworn that higher numbers meant more aggressive. I will have to do some testing with it. I need to read about and try and understand drivers2.txt more.

Keep in mind that the carset you build will not be a perfect recreation of historic results, but rather an approximation of the possibilities/probabilities of what could have happened. Your top three drivers have won 70% of the races you've run, which seems pretty good.

I know what you mean. My goal is to make the performance basically mirror what we saw in real life while making it slightly more competitive. For 1996, this would mean that the Ganassi cars won't be so completely unstoppable Guys like Al Unser Jr need to be able to win, and Bobby Rahal as well. (ever now and then- not as strong as the honda's were though).

When I re-watch the races from '96 there were a bunch of situations where competitive drivers got knocked out and "should" have won. I try and capture this in drivers2.txt (think of AUJ blowing an engine in the lead on the last lap at Elkhart lake!)

[CowtownBob]

I was surprised about the aggression being negative, but here's how I determined it: I made up a drivers2.txt where all other values were equal, then gave each driver a discrete aggression value so that the field covered the whole range from 0 to 999. Then ran somewhere around 20 races (don't recall exact number). Cars with low values tended to move up through the field more than those with high values.

[checkpoint10]

This is an interesting topic and I didn't realize that aggression might be negative. I performed CowtownBob's experiment, but I found that, all other things being equal (I had forced all the power/traction/drag to be almost equal by setting them to 500-501), the "Zero aggression" drivers always started ahead of the "999 aggression" drivers. Because of this, my Zero drivers always did better in the race, but I couldn't tell whether that was because of where they qualified. I did another test where I attempted to minimize that issue by assigning everyone 1-999 for their power/traction/drag, and I believe I noticed the effect that CowtownBob described - those drivers with neutral or zero aggression tended to finish higher than they qualified - but I didn't formally keep track of the results so this is just a hypothesis at the moment.

[checkpoint10]

As a follow up to my earlier post, I just ran 10 races of 20 laps each at Phoenix, with a field of 24 cars split evenly between "Neutral Aggression" (500-501), "Zero Aggression" (0-1) and "999 Aggression" (998-999). All of them had power/traction/drag range of 400-600 for randomness in grid spots, and the chassis/engine/tire at 0/0/0. I chose Phoenix because I thought the field would keep bunched up during the race.

The results are as follows:

Average starting position:
Zero: 11.74
Neutral: 12.61
999: 13.15

Average finishing position:
Zero: 10.25
Neutral: 12.44
999: 14.81

Average position change:
Zero: +1.49
Neutral: +0.18
999: -1.66

This suggests that higher aggression values make the driver slower in qualifying as well as more likely to drop places during a race. If this is true, I may have to revise the drivers2.txt for my carset because my understanding was backwards. Finally, I'm no scientist or statistician, so please critique my methodology.

[samsepi0l]

Finally, I'm no scientist or statistician, so please critique my methodology.

Your methodology seems pretty damn scientific to me! Good job. I have to revise a few of my carsets as well.

[checkpoint10]

I just did another experiment to figure out the pace of the AI when using different chassis/engines/tires. I set it up as follows:
1) Performance values are all 499-500.
2) 3 lap races, standing start, only one opponent, constant 70 degree weather, no wind
3) As soon as the race begins, I turn on accelerated time and record the final average speed and elapsed time for the single opponent (winning car). The idea is that the AI can put in the laps without any disturbance from other AI or myself.

Unfortunately the results may suggest that it doesn't matter what the AI is driving:

Milwaukee test:
Lola-Cosworth-Goodyear made 3 laps at Milwaukee in 66.9 sec, or 166.406 mph. I can repeat this exactly three times with the same car. Also it is same result if I keep my car on track without accelerating time (in case for some reason the AI drives differently under accelerated time). But then, I gave the AI a Lola-Mercedes-Goodyear and it produced identical times as the Cosworth!

Elkhart Lake test:
So I moved to Elkhart Lake, thinking maybe the length of the course would amplify any differences. I tested the following: Reynard-Honda-Firestone, Lola-Mercedes-Goodyear, Penske-Cosworth-Goodyear, using the same methodology as above. All three combinations produced the same 5min 19.7s time, 138.106 mph average.

My conclusion is that, given a clear track, the AI will produce the same lap times regardless of the car combination. (I'd love to be proven wrong, please again let me know if there is something else that might help show any differences!)

I am now curious whether there are differences in the player's car. I wonder if there is a reliable way to test this.

[samsepi0l]

I tested a 100% 83 lap race at midohio with accelerated time, and all drivers with aggression min = 350 and max = 950. It seemed much more competitive. I am going to run this in my drivers2.txt for a while and see how I like it.

[kakukri]

I am now curious whether there are differences in the player's car. I wonder if there is a reliable way to test this.

There are differences in the player, I could feel it when I was regularly playing ICR2. Also the tire temperatures were different, on ovals the Lola had about 10 F lower front tire temperatures (at least with that particular setup I have tested). Many other drivers felt the difference, basically the Lola is a bit understeery, the Penske neutral and good to change direction, the Reynard has an oversteering tendency, the Ford has better low rev torque, the Honda has better top power, for the Mercedes I do not remember. The differences are not that big, though.

As for the AI cars, considering the computer power available in those days, it makes sense that they have a simplified physics model, not the same as the human player.