A team of boffins from Intel Labs and Darmstadt University in Germany have come up with a cracking way of playing computer games on work time -- trying to see if machines can be taught to drive using simulators like the Grand Theft Auto series.
It's not so much about the controls and being able to successfully perform a doughnut, with the machines more interested in seeing the environment and practising to see if they can work out what's a pavement, what's a road, and what happens when one of those two-legged things are hit at high speed.
Seeing as the most recent Grand Theft Auto is bordering on the photorealistic in some places, and packed with stuttering AI cars that replicate the confused driving of humans, it's a good test track for machine learner drivers.
The team inserted a software layer between the game's world and the machine driver, letting it classify roadside objects all by itself, before attempting the terror of the reverse around a corner. [MIT via PopSci]