AC - AI Level

[Leigh Nielsen]

Can AC AI level be set beyond 100% - e.g., like AMS.

It could be useful to let newer drivers know that to generally keep up with the H2H field they should be able to lap at X% in SP for a given car/track combo. With some combinations it is relatively easy to beat the AI at 100% while others not so.

 

[BhZ]

Can AC AI level be set beyond 100% - e.g., like AMS.

It could be useful to let newer drivers know that to generally keep up with the H2H field they should be able to lap at X% in SP for a given car/track combo. With some combinations it is relatively easy to beat the AI at 100% while others not so.

No it's not possible. Just use RSR to see the average laptime of the drivers.

 

[kolysion]

of course you can, AI.ini, set from 0.00 to 1.00 AI at 1.2

 

[Stereo]

AI at 100% is just as fast as the AI knows how to drive, to go faster it'd have to stop obeying the laws of physics which doesn't really make for good competition. If you do edit an ini file above 100% they'll just crash as soon as they try to take a turn cause they'll think it can have more than 100% lateral grip.

If you want them faster on a track you'll have to record your own AI line that they can drive quicker. They use a generic one so doing one for the type of cars you want to use can be effective.

 

[ltcars]

AI at 100% is just as fast as the AI knows how to drive, to go faster it'd have to stop obeying the laws of physics which doesn't really make for good competition. If you do edit an ini file above 100% they'll just crash as soon as they try to take a turn cause they'll think it can have more than 100% lateral grip.

If you want them faster on a track you'll have to record your own AI line that they can drive quicker. They use a generic one so doing one for the type of cars you want to use can be effective.

You can go to more than 100%, I think Kunos builds in little safety room, I would not go over 120%

 

[ltcars]

You can go to more than 100%, I think Kunos builds in little safety room, I would not go over 120%

http://www.racedepartment.com/downloads/faster-ai-for-all-tracks.14758/

 

[Giovaneveterano]

With the multiplier over 1 the AI brakes later into turns, sometimes it can slow them down, sometimes it makes them faster.
I suggest to play on the value but not every track/car combo can be faster than 100%.

Let the AI running some qualifying/practice laps before the race makes them way faster than jump straight into the race. For me this last one is the best solution

 

[Insert Coin]

With the multiplier over 1 the AI brakes later into turns, sometimes it can slow them down, sometimes it makes them faster.
I suggest to play on the value but not every track/car combo can be faster than 100%.

Let the AI running some qualifying/practice laps before the race makes them way faster than jump straight into the race. For me this last one is the best solution

Serious question: they learn?

 

[ltcars]

Serious question: they learn?

Yes, you can see it when in early stages of the practice they miss their braking points, lock up etc. It is all lost after the raceweekend

 

[Giovaneveterano]

Serious question: they learn?

Sure. They adjust the setup.

 

[Neilski]

Holy smoke, who knew!?
(But seriously, who knew? I guess this is written down somewhere?)
I had no clue that the AI learned but then again I hardly ever race against them...

 

[AccAkut]

http://www.racedepartment.com/downloads/faster-ai-for-all-tracks.14758/

IMO that is something different, the AI of the car will still only work at 100%, but you basically tell it the wrong limits, like putting a 100km/h sign in a 30 zone.

Holy smoke, who knew!?
(But seriously, who knew? I guess this is written down somewhere?)
I had no clue that the AI learned but then again I hardly ever race against them...

There was talk with Stefano about this, if these value should be stored in a way, but iirc he concluded that the parameters of each session are so different each time that storing these would make the AI more unpredictable, but not faster.

 

[Neilski]

should be stored in a way

Ah - when I said "written down somewhere" I meant that the learning ability is documented perhaps? I had no clue it even existed. I can see the argument for not making the learning persistent - the mere fact that they can learn at all is pretty cool.

 

[Majuh]

It's wrong, they don't learn: http://www.assettocorsa.net/forum/index.php?threads/race-weekend-ai.34222/

 

[AccAkut]

It's wrong, they don't learn: http://www.assettocorsa.net/forum/index.php?threads/race-weekend-ai.34222/

eh, that post there is from 2016, long before the big AI updates, and has no word from the Dev at all. Its down to what you define as learning, but if "they don't learn" means they drive the exact same the whole session, well you're wrong. Start a Qualifying session and take a look in the log.txt while the session is going on, you will find stuff like this:

AI CAR:9 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
Setup change for Car: rac [9] Changing: WING_2 from 6.000000 to 4.000000
Setup change for Car: krb_tiga_lotus [0] Changing: FRONT_BIAS from 0.600000 to 0.600000
Setup change for Car: bmw_m3_e30_dtm [12] Changing: WING_1 from 3.000000 to 2.000000
Setup change for Car: ks_audi_r8_lms_2016 [1] Changing: WING_1 from 2.000000 to 1.000000
Setup change for Car: ks_audi_r8_lms_2016 [1] Changing: WING_2 from 6.000000 to 7.000000
Setup change for Car: bmw_m3_e30_dtm [13] Changing: WING_1 from 3.000000 to 2.000000
AI CAR:14 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
Setup change for Car: ks_mercedes_amg_gt3 [2] Changing: WING_1 from 2.000000 to 1.000000
Setup change for Car: ks_mercedes_amg_gt3 [2] Changing: WING_2 from 6.000000 to 7.000000
Setup change for Car: ks_nissan_gtr_gt3 [3] Changing: WING_2 from 9.000000 to 10.000000
AI CAR:4 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
Setup change for Car: ks_audi_tt_vln [5] Changing: FRONT_BIAS from 0.830000 to 0.900000
Setup change for Car: ks_audi_tt_vln [6] Changing: FRONT_BIAS from 0.830000 to 0.900000
Setup change for Car: ks_porsche_911_gt3_cup_2017 [7] Changing: WING_2 from 8.000000 to 7.000000
Setup change for Car: renault_megane_trophy [8] Changing: WING_2 from 8.000000 to 6.000000
AI CAR:13 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
AI CAR:3 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
AI CAR:8 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : M
AI CAR:2 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
AI CAR:12 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
And the winner is 0 : S
AI CAR:7 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : H
AI CAR:11 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
AI CAR:1 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : M
And the winner is 0 : S
AI CAR:6 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S
AI CAR:10 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : M
AI CAR:15 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : M
AI CAR:5 looking for qualify tyre, lap length: 4189.7 REQUEST:10474.2
And the winner is 0 : S

This is right at the start of the session, AI does preliminary changes to the stock setup before they exit the pits for the first time

CAR 5 RETIRED Session 2 : LC:1\0
CAR: 5  setting fuel for 6 laps, with mult=1.100000
Using RaceEnginner,Car::setRequestedFuel: 13.555833
 fuelxlap=2.053914 fuel 13.555833
CAR: 6  setting fuel for 6 laps, with mult=1.100000
Using RaceEnginner,Car::setRequestedFuel: 13.555833
 fuelxlap=2.053914 fuel 13.555833

this happened a few minutes later, AI does adjust fuel. On other track I have seen enrties of them adjusting wings or diff ratios (this was a short test at Zandvoort). This by my definition is an AI "learning" during a session.

 

[Majuh]

eh, that post there is from 2016, long before the big AI updates, and has no word from the Dev at all. Its down to what you define as learning, but if "they don't learn" means they drive the exact same the whole session, well you're wrong. Start a Qualifying session and take a look in the log.txt while the session is going on, you will find stuff like this:

This is right at the start of the session, AI does preliminary changes to the stock setup before they exit the pits for the first time

this happened a few minutes later, AI does adjust fuel. On other track I have seen enrties of them adjusting wings or diff ratios (this was a short test at Zandvoort). This by my definition is an AI "learning" during a session.

No... The changes to the wings are entirely random and not based on any observations made by the AI. Same thing goes for the fuel load, there is no meaning behind it. Just watch the AI in qualifying and you will see that they simply use a random fuel load each time they exit the pits.