Alternatively Intelligent?

This is a submission from TSA member, JamboGT and is going to be used in his research for Uni.

Artificial intelligence is a topic that covers all types of game. However, unsurprisingly for those that know me, I want to talk about artificial intelligence in racing games. I study computer programming and game design and am currently working on my final year project, a form of adaptive AI in a racing game.

When making a racing game, I feel that one thing that is very tricky for developers is balancing the AI opposition so that players of all ability can not only defeat the opposition but be able to have a compelling and exciting race with them. Some players may be much faster than others so what do the developers aim for? The players running perfect laps or the players who make the odd mistake and go a lot slower? For many games the answer to this problem has been a technique known as rubber-banding.

For those of you that don’t know, rubber-banding is a method that artificially increases or decreases the performance of the AI opposition dynamically in order to match that of the player. This may be done by increasing the speed and acceleration of the AI, increasing the grip level of the AI or, in weapons-based games like Mario Kart, a better chance of picking up a good weapon.

Rubber-banding, for some games, is a great solution. It keeps the race close the whole way until the chequered flag. Would Mario Kart be Mario Kart without Luigi having a run at you on the final lap?! Burnout wouldn’t keep the same thrill until you crossed the line if the AI wasn’t using rubber-banding to catch up following a crash or a great burnout by the player. However, while this may keep the excitement levels up, there is always the feeling that the AI have been cheating, that the game is skewed unfairly against the player.

Sim games, such as Forza and Gran Turismo, tend not to use rubber-banding. GT has a difficulty that scales as the player levels up and Forza has a difficulty setting. There is no adapting of the AI as you race. This may lead to a “purer” race but also has its own problems. Namely, finding a skill level relative to that of the player. To put it simply: some people are good at racing games, some people aren’t, yet the AI has to race against a wide spectrum of abilities. This means that for some players they are far too fast and for some ridiculously slow.

For my final year project I decided to try to find a solution to this problem. To find a type of AI that can adapt to different players’ abilities dynamically yet not do it in a way that seems unfair to the player. A basic explanation is a system that tracks the players’ performance over a series of races and laps and then, over time, adjusts the AI to be able to provide a competitive opponent.

To do this, I need your help. Just down there is a link to a questionnaire (a very quick and painless questionnaire!) that I will be using to aid my research, the more data I have, the better my research so please take the time to fill it out.

Please click here to take the questionnaire!

– PAGE CONTINUES BELOW –

81 Comments

  1. Done! Hope it helps! :)

  2. All done. Very interesting, there are times when I love the feeling of pissing off into the distance and other times fighting to the line, it just depends on my mood. I always start a new racer on the easiest setting, if it has one, and quickly build up to a harder setting. Absolutely love racers.

  3. Thanks to everyone for the responses so far, it is very much appreciated!

    I will be following up as much as possible, there may even be graphs!

  4. Done! Good luck with this. I’m expecting this system in GT6 ;)

  5. Done. And curious. Is my memory playing up or did GT4s B-Spec mode not have a system in place that analysed your driving and used that information to train your B-spec driver?

  6. All done, for the sort of games I really enjoy rubber banding is very important.

  7. My favorite IA in a racing game is Toca Race Driver 3. There are a lot of opponents on the track (like 20 or so), each with different skill level. So no matter what skill level you have, np matter what silly accident may happen to you in a race, you still have opponents to race against. But there’s o hiding that if you make a mistake you’re almost sure not to catch up to the best racers. So you have a challenge throughout the race (overcoming the next opponent), and there are real stakes for your mistakes. It doesn’t matter if I don’t win each time, but I have been able to confront myself with lots of opponents, and find myslf naturally competing with opponents of similar skill.
    AI then should just make sure opponents sometimes make mistakes, or allow other to overcome to keep things interesting and fair.

    • That’s pretty much what I outlined briefly in the survey. I’ve never played Toca so I didn’t know something like this could be found in a racing game already. Most racing games have all the racers bunched up together. If you’re not within that few seconds gap, you’re racing alone either in front or behind them. BORING.

    • I think my favorite so far is another Codies game, Indycar Racing, here is an article on it http://uk.gamespot.com/news/6023395.html

      This was a system that trained the AI and evolved them over time.

  8. Done, good luck with it and a great topic I’d never really considered.

  9. I’ll do the questionnaire in a short while.

    Of interest to you though may be the ‘Director’ in the left 4 dead games, not racing, but their approach and intentions when coding it may be useful.

    • Oh and Burnout 3 for example was exciting but damn annoying if you had a perfect race only to crash at the last minute because you WOULD finish last.

    • One of my housemates has been studying this for a different project, it’s very interesting. A good shout there.

      • The L4D system works really well in it’s own little genre, and I believe the effect that The Director has in the game is what makes it still so loved long after release.

        The issue I have with the rubber-banding difficulty level is that AI drivers seem to lose their margin of error. It might make it exciting to see them gaining on you, but the extremities of their perfection can be annoying. A better system would be to have them technically superior to you due to the chase-and-adrenaline effect, but have a higher chance of error due to their pushing the perceived boundaries of what is possible.

  10. Did it Jambo , played them all except Split Second . After Burnout Paradise i realised i strongly dislike arcade racers nowadays.

Comments are now closed for this post.