Google's Testing Self-Driving Cars In a Matrix-Style Simulation

Illustration for article titled Googles Testing Self-Driving Cars In a Matrix-Style Simulation

Somewhere in the Googleplex, Google has a digital simulation of the entire Californian road system to test its self-driving cars. This, says Google, allows it to perform "decades worth of testing" in just hours.

Advertisement

Indeed, Google has such confidence in its system that is now lobbying the California's government to certify these self-driving vehicles based solely on these virtual tests rather than real driving. Rob Medford, Google's safety director for the self-driving car has written a letter in early 2014 to California state officials, which The Guardian has obtained. Medford writes: "Computer simulations are actually more valuable, as they allow manufacturers to test their software under far more conditions and stresses than could possibly be achieved on a test track."

That's all fine, but the DMV still isn't biting. Not only does it still want Google to still test its cars in physical conditions, it has also put new rules in place requiring Google to put in backup steering wheel and pedal system in the cars. Just in case.

Advertisement

The Guardian reports that the Google cars have so far virtually driven more than 4 million miles inside the Matrix-like simulation, facing real-world challenges like lane-weaving motorists, wobbly cyclists and unpredictable pedestrians. Meanwhile, in real life Google's cars were last said to have driven around 700,000 miles over a road system that's reportedly just 2,000 miles long. [The Guardian via The Verge]

Share This Story

Get our newsletter

DISCUSSION

sdfkljsbafhvfda
sonicsight

Gotta agree with the DMV here. Virtual testing has it merits, that's undeniable. I think its great they have this simulation to test these thing. It allows them to thoroughly and extensively test most scenarios. But the problem with virtual testing is it can only test the scenarios that the developers think up. That is a very limited subset of the possible scenarios that EVERYONE could think up. They have to be tested against what any random person could end up doing in the real world, you simply can't account for that with a virtual simulation (at least not until we figure out true AI). So its great that they do this, but it should be an addition to real world tests, not a replacement for them.