• Register
Post tutorial RSS Training the AI

This article contains information on training LFS's Artificial Intelligence (AI) system so that your Non-Human opponents can get harder and trickier to beat.

Posted by on - Basic Other

The AI (Artificial Intelligence) in LFS is very good considering the AI in other Sims, the AI learn as they drive and compete, however the lines they use are quite weird, It appears at slower speed and corners the lines are not bad, but as levels and speed increase, they bounce off walls and do some stupid things.

To train the AI you select the level of learning and setup, you want to use and set the track and amount of laps. The AI learns in all modes as long as 1 lap is completed, It is suggested to be a good idea to include a faster and a slower car in the pack to randomize the pack and make the AI work to pass slow etc. Some suggestions for future updates to training have been taken onboard, such as a progressive training mode that will cycle through a set range of tracks cars and learning levels adding faster and slower cars, so you can set and train overnight / weekends etc in unattended mode.

There are five different levels of training availavle from newbie to pro. It appears that there is no cumulative effect or progressive knowledge gained from a previous level, this is based on each .knw file being specific to a track and level, to explain this further, we’ll look into the use of the .knw file, This is found in the knw folder here the stored results of the training for each individual AI is kept. I.E. You train AI 1 at South City Classic in “learner” mode using FXO TURBO car, it saves the completed knowledge file (.knw) as “AI1_FXOTURBO_SthCty_Clssc_LEARNER.knw”, now if we train the same AI1 and car / setup at the higher Pro level it saves a different file “AI1_FXOTURBO_SthCty_Clssc_PRO.knw”. This indicates there is no progressive learning, which is a shame I’d like to see the AI learn from previous levels incrementally, at lower/slower levels the AI have smooth non-wall banging moves, this disappears the moment Quick or Pro levels are introduced. Beware if at a later date you change the AI name from AI x to a proper name like Schumacher or Brock, it will no longer be trained, it depends on the .knw file to be the same name as the AI or that training is not associated with the car, The same doesn’t apply to setups, change the setup with the same AI and it adds that setup to its knowledge base, this is evident when the race starts as there is no “drivers first time” message displayed. The best way to train the AI is to be methodical about it. First determine what level of competence you really are at in LFS in your own driving ability, be honest! Then train the AI at that level to provide you with some good competition and improve your skills.

These are the best ways I find to train at first, select your range of cars, I select 8 of one class, maybe gti and add some faster cars and make one GTi real slow by fooling around with the gearing so it can only do 80 mph top! Set the laps to practice, track to whatever and start racing, record the times of the main cars (GTi class) after first flying lap is completed (end of lap 2) and go to bed or away for the weekend. I have made an Excel worksheet for this purpose (Avail at RSC). When you return, record the last lap times in the worksheet I have provided, giving you the improvement time for that car at this level of training. Then end race.

This saves the .knw files, repeat for all cars all tracks all levels and then race against the AI at the lower levels till you can pass them under brakes avoid the mishaps and beat them every time, then you move up to the next level of AI, when you can race and beat them at Pro level go online, you are now experienced and safe to race against. Scawen suggests the AI learn to a high level of competence in about 60 laps.


This comment is currently awaiting admin approval, join now to view.

Post a comment
Sign in or join with:

Only registered members can share their thoughts. So come on! Join the community today (totally free - or sign in with your social account on the right) and join in the conversation.