Monday 29th August 2016
Strict laws in the US have forced Google to rethink how it develops autonomous car technology in the future
The much-hyped self-driving Google Car may have reached the end of the road, with its technology boss Chris Urmson, who led the project for seven years, resigning at the beginning of August.
The departure comes amid a move from Google to team up with the automotive industry, rather than challenge it, in the race for autonomy.
According to a note sent out by auto analysts at investment bank Evercore ISI, the Google car project was clearly in trouble. “The Google driverless project is likely not as advanced as many believe,” it said.
Speaking at Austin’s Southby-Southwest conference earlier this year about the timing of the project, Urmson said: “How quickly can we get this into people’s hands? If you read the papers, you see maybe it’s three years, maybe it’s 30 years. And I am here to tell you that honestly, it’s a bit of both.”
Perhaps one of the reasons for Urmson’s exit was the Google Car project’s philosophical U-turn performed last September, with the hiring of ex-Hyundai US boss John Krafcik as CEO of the Google self-driving project. Google has also just signed a deal with Fiat-Chrysler to build autonomous prototypes based on the company’s Chrysler Pacifica MPV, all of which suggests that any idea that Google would put its own self-driving cars into production now looks increasingly unlikely.
Perhaps such a tie-up with the car industry was inevitable. Despite the time spent on testing and development, the Google self-driving cars had covered a relatively modest 1.7m miles as of this summer. In stark contrast, Tesla claimed that, by May this year, drivers of its cars had clocked up 100 million miles with the Autopilot function engaged. While this is nowhere near the same as the intended full autonomy of the Google cars, it shows the huge disadvantage of trying to jump straight to a fully autonomous car from experimental vehicles.
Furthermore, not only do the Google cars have a selfimposed 25mph top speed, it took until 2011 for fully autonomous testing on public roads to become legal in the US, when Nevada changed its local laws. California — Google’s home state — didn’t change the law until this year. It is still only legal in four states.
However, California’s regulations contained one crucial clause, which could well have sunk the Google Car for good. Under Article 3.7, paragraph 227.18 of the order, it states: “A manufacturer shall not permit any of its autonomous vehicles to be operated on public roads in California when the operator is not seated in the driver’s seat and either monitoring its operations and able to take over physical control of the vehicle, or in physical control of the vehicle.”
The requirement for autonomous vehicles to have a conventional steering wheel and brake pedals was a major setback for a stand-alone Google car. It’s understood that Urmson was involved in political lobbying to try to get this clause rescinded. His departure makes this unlikely, especially as major car manufacturers are already testing conventional cars fitted with autonomous tech, such as Audi’s A5 and A7 prototypes.
But despite these departures and legislative hurdles, the Google project remains committed to seeing true self-serving vehicles, according to a recent Bloomberg interview with new project boss Krafcik.
He highlighted the recent fatal accident involving a Tesla driver using the Autopilot function as an example of the problems with what’s known as ‘level two’ autonomy.
Krafcik revealed that Google tested this combination of radar cruise control and lane keep assist in 2012 and found that its drivers started to zone out and even started texting or “reaching into the rear seats”.
Krafcik says he remains convinced that full level four autonomy, “which is our focus at Google”, is still the best outcome. He also referenced the onerous new Californian regulations, which would require the driver of a future autonomous vehicle to have a second driving licence for operating a self-driving vehicle, something he says he’s against. Krafcik said that Google’s deal with Fiat Chrysler Automobiles calls for 100 autonomous Pacifica prototypes, “which would double our test fleet”.
Tellingly, he also admits that building a car is a significant hurdle. “Google realised that it’s really hard to build a car,” he said. “We built the little prototype that gave us a taste of the complexity.”
So, while the stand-alone Google Car may be dead, the project to embed Google self-driving technology into future models from mass car makers clearly isn’t.
Google autonomous vehicle crash - are we ready for self-driving cars?
Fiat Chrysler Automobiles to test autonomous cars with Google
Government launches major consultation on the future of autonomous cars
Dramatic changes for UK motoring predicted with autonomous cars
Are Google Maps good enough?
Another big issue for the Google Car project is whether Google’s own mapping base can be used for autonomous driving. Many in the industry say it can’t, which is why the HERE mapping division was bought from Nokia by Daimler, Audi and BMW.
Google maps are, in effect, a kind of electronic 2D A-Z map. A device running the maps application relies on GPS satellites to ‘dead reckon’ its position on the road. This isn’t easy, because the civilian GPS network is accurate to only 15 metres.
Although the system has been honed to work very well as a conventional navigation set-up, it is nowhere near intelligent enough to be used for autonomous driving. It cannot place the car on the road with anything like the required accuracy and it cannot work with forward-facing cameras to read the road ahead, identifying potential immobile obstacles such as roadworks and temporary traffic lights. That’s why the Google test car were equipped with Lidar — a kind of laser-based radar system — for reading the road ahead in 3D.
The HERE mapping network is a huge step forward over Google maps. The company has used Lidar to map most of western Europe as a kind of 3D ‘cave’. Not only are the roads mapped, but also the pavements and building frontages. Once a HERE-equipped car is inside this ‘cave’, it can be positioned to an accuracy of just 10cm — ideal for autonomy.