Google, it has emerged, has built a “Matrix-style” simulation of the entirety of California to test its self-driving cars. While this in itself isn’t massively surprising given Google’s history as a software company (though it is a bit scary), the company is also petitioning California’s state officials to allow safety testing within the Matrix, instead of testing on real roads. This might sound a little terrifying — imagine if Ford started selling a car that had never been road-tested — but it makes quite a lot of sense for a self-driving car, where there are millions of conditions that need to be tested — conditions that are essentially impossible to test in real-time on real roads. “In a few hours, we can test thousands upon thousands of scenarios which in terms of driving all over again might take decades,” a Google spokesperson told the Guardian.
Information about Google’s Matrix-like simulation of California was obtained by the Guardian – via a freedom of information petition to California officials, and then some further information from a Google spokesperson. Google has built the entirety of California’s road system (about 172,000 miles) in software, along with accurate simulations of traffic, pedestrians, weather, and so on. There’s no word on the hardware being used to create the Google Matrix, but it’s probably a fairly large cluster of servers.
Google’s virtual self-driving cars have so far driven more than 4 million miles within the Californian Matrix, facing all of the usual challenges that its real-world self-driving cars might face (wobbly cyclists, vehicles running a stop sign, etc.) By comparison, Google’s physical fleet of self-driving cars (mostly modified Toyota Priuses) had only driven 700,000 miles as of April 2014 — and more importantly, they have only driven on around 2,000 miles of road.
Clearly, especially when it comes to self-driving cars, detailed simulations can provide even more feedback than real-world testing. Currently, most countries require real-world testing for new cars — on test tracks, closed public roads, etc. In California, self-driving cars are beholden to the same regulations: They must be road-tested.
Google, however, wants California to change that policy, to allow for self-driving cars that have only been tested in the Matrix. Earlier in the year, according to the Guardian’s freedom of information request, Google wrote the following to California state officials: “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 … Google wants to ensure that [the regulation] is interpreted to allow manufacturers to satisfy this requirement through computer-generated simulations.”
While I’m all for Google carrying out exceedingly detailed simulations — and I think it’s the only way that Google can bring a fully autonomous car to market within a reasonable time frame — I think some level of real-world testing is probably a prudent idea. You can make a simulation as detailed as you want, but it can never quite capture the full gamut of weird and wonderful things that a brain-powered pedestrian or fellow road user is capable of.
If Google wants to put vehicles on the road that are completely controlled by an on-board computer, and wants to avoid the tsunami of lawsuits that will surely follow, it will need to really prove beyond doubt that its self-driving AI is comparable to a human driver. This is a very different approach from conventional car makers, which are coming from the opposite (and much safer) direction — taking a normal car and slowly adding self-driving-like features. There are still some serious questions about how we handle the ethics and morality of robots — and an autonomous self-driving car is just one particular breed of robot. I think we are quite a few years away from building a fully autonomous car that can make such decisions as “do I run over the cat, swerve into a tree, or brake hard and cause a pile-up.”
In other news, California also recently ruled [PDF] that any self-driving cars must be fitted with a backup steering wheel, for situations where “immediate physical control” is required. This is a bit of a hit to Google’s new self-driving car prototype, which eschews the standard car controls for a single “Go” button.
Author: By Sebastian Anthony