The new method is probably meant more for modders and track builders to create better AI rather than for average user. I found the instructions quite clear. Although ISI could quite easily expand this functionality e.g. like this:
Save AIW file after learning procedure into UserData/player/Settings/*trackname* folder. Then at each track loading rF2 would check if there is an updated AIW in UserData and override the track default with the user one. No need to repackage track then.
I am a track modder and I also found it easy to "make it work" but that doesn't mean that it was correctly explained. There are many important things that are not explained (I will expand on this later). You also suggest the functionality is for modders and track builders and also mention an updated AIW. The fact that you dont need to repack, suggests that the tool is meant more for casual racers that don't know how to improve the AIW using Dev Mode. If there was an updated AIW file, it could actually help modders. However, none of the 3 autocalibration methods does this.
AC1:
ini A file that provides some parameter per waypoint of the fast path with this format "lat=(678,0.063,-0.000)". It is not explained what the middle value is.
AC2:
wis The stupid of me (the method is probably perfectly explained) always get empty 2KB files when trying to use this method so I don't know its format
AC3:
tl3 A file that provides some parameter per waypoint of the fast path (like the first one) with this format "int=(0,1.016,0.000)". It is not explained what the middle value is.
The parameters of methods 1 and 3 seem to affect the wp_pathinfo2 parameter of the AIW that describes the fast path (0). Looking at the values in .ini and .tl3 the first one seems to provide a value to be added to the original wp_pathinfo2 parameter and the second one seems to provide a value to multiply the original parameter. If this was confirmed, it would be really easy to adapt an AIW file with the newer using an excel sheet.
Actually - from reading marcG's post they offered 4 lines of explanation and Marc wanted a clarification. Had they been perfect you would have been left with one less thing to scold them for which may have created a huge void in your life....
It is really interesting the way people defend the undefendable in this forum. The new learning method has been as poorly explained as they were the other first two learning methods. If you would take a look in older threads you would realize that people have always had trouble with this. If everything was as clear as you suggest ther wouldn't be threads like this:
http://isiforums.net/f/showthread.php/21197-Autocalibrate-AI-TIM-ISI-input-needed-please
http://isiforums.net/f/showthread.php/21073-AI-track-learning-mode
BTW
A good manual (as I understand it) would at least include the following notes and clarifications:
- A good description of how to activate each learning mode.
At the moment this is poorly done. I still cannot get AC mode 2 working properly. I am not sure how to specify which
- A recommendation of the required number of laps to get a significant improvement for methods 2 and 3.
- An explanation of how the AI and the track should be setup in order to get better results: AI strength, AI aggression AI limiter, Amount of Rubber (green or saturated). I am not sure how these parameter affect the result of the learning.
- A clear explanation of how to use it afterwards. Does the game automatically use learning files if present? or should we specificy autocalibrate mode in player.json? What happens if there is one file of each type? Does some method have priority over the others? How can modders include the tl3 in their tracks? should they specify one per each track/car combo?
Besides of that, there are IMO several contradictions in this whole autocalibrate system:
It is not clear what is the main purpose of it. Is it for modders or for users?
I am not sure either if it is meant to improve a poor AIW so that some tracks are playable offline or, on the other hand, if it is simply thought to adapt a good generic AIW to a specific car.
In any case, it is quite funny that in 2 of the methods the result only applies to one specific car within a mod (for example: car#22) considering that the AIW file is generic for all cars, it would be nice if the results of a learning method could be applied in general as well. Apparently AC mode 3 is more logical in this sense and applies to all the cars of the same type which is not so pissing to extend.
I have tried to use many times autocalibrate in the hope to use the AI in dev mode so that they could define a better AIW. However, it is impossible to do properly since the AIW is applied immediately to the car when the best time is beated instead of only after finishing the stint. This provokes the AI doing weird things. I would think that the refinement should come after picking the best lap of each stint. Furthermore, I am not sure if the best lap time is the best criteria for this. In general the fastest time may come for being very close to the limits and this would provoke a very inconsistent AI.
