New AI Challenge Is All About Wanton Destruction 45
togelius writes "Previous years have seen a number of car racing competitions where neural nets, rule-based systems and other fancy AI techniques have been put to the test by letting them drive on a track and seeing who gets the best lap time. Recognizing that finding the Michael Schumacher of AI is not enough, a team of researchers from University of Wuerzburg now wants to find the Mad Max of AI. Their new competition is called 'Demolition Derby' and the goal is to 'wreck all opponent cars by crashing into them without getting wrecked yourself.' For this, they use the open-source TORCS game and a custom AI interface, allowing all and any AI researchers and enthusiasts (including you!) to submit their best and most aggressive controllers." (There's a competition for conventional racing, too; competitors can enter either or both.)
The rules are pure idiocy (Score:4, Insightful)
All of these rules are the opposite of how actual demolition derbies work. Smashing a car into the wall causes large amounts of damage. Damage to the front (the radiator, engine etc) is way more effective than damage than the rear. Most cars are driven backwards because the trunk is just a big crumple zone. As long as it doesn't get the wheels or axle, damage in the rear doesn't really matter.
OK, they need to call it something else now. It's not a demolition derby anymore. There's no floating wrenches that repair all damage in real life either.
Re:The rules are pure idiocy (Score:5, Insightful)
Perhaps the goal is not so much about creating an accurate demolition derby simulator, so much as it is about creating a suitable challenge to improve artificial intelligence? ... you know, like how chess is a terribly unrealistic war simulator, while still being useful in training strategical thinking.