Teslas Self-Driving Car Plan Seems Insane, But It Just Might Work
Elon Musk has done it again. The CEO of Tesla Motors and SpaceXhas made an aggressive prediction for his technology, and paired it with an unreasonable deadline. In the past, he’s promised electric cars for everyone and trips to Mars, and built a reputation for achieving thoseoutrageous things, albeitfar behind schedule.
Now, Musk is pledging that by the end of 2017, he’ll produce a Tesla that can drive itself from Los Angeles to New York City, no human needed.
That timeline puts himyears ahead of everyother big player working on fully autonomous cars.Ford is aiming for 2021, China’s Baidu for 2019. Google hasn’t given a hard date, but 2021 is a good bet. Same goes for GM. And they’re talking about cars that will stay in limited areas—nothing that can cross the country.
So, what are the chances Musk can actually make this happen? Let’s break it down.
Tesla’s website already provides a checkbox for Full Self-Driving Capability,” claiming, all you will need to do is get in and tell your car where to go.” But the cars won’t actually do that, at least not for a while. Musk isn’t issuing the necessary software. These earliestcars won’t even have first generation Autopilot, Tesla’s system that keeps the vehicle in its lane and a safe distance from other vehicles on the highway.
The automaker did release a video of its Model X SUV, carrying the new hardware and prototype software, driving itself around Palo Alto. It seems impressive, but it’s no proof of concept.
“An unedited sequence of the vehicle driving itself in downtown San Francisco would be more meaningful to see what the vehicle is capable of doing,” says Raj Rajkumar, who runs autonomous driving research at Carnegie Mellon University.
Let’s say the car’s as capable as the video makes it seem. Could it get itself across the country? “If were looking at another year of development, I think we could see something like that happen, says Jeffrey Miller, an IEEE member who studies autonomous driving at the University of Southern California.
Yet Rajkumar points to Google, which, after nine years of work on this project, still won’t even give a timeline for its rollout. “Mastery of self-driving under real-world conditions is not going to be easy,” he says.
And you have to consider Musk’s track record of missing his own deadlines. Tesla products that have missed their original due dates include the Roadster, Model S, Model X, Model 3, and Autopilot. So, kinda everything.
Verdict: It’s plausible Musk will meetthis deadline, but his own track record says otherwise.
Building a car that can handle any situation requires a lot of teaching. Google is the acknowledged leader here: Its fleet of 60 self-driving cars has covered more than two million miles, mostly around a few cities. But Tesla claims an advantage: Its cars have covered over 222million miles in Autopilot mode, all over the world, collecting data the whole time.
“Theyre getting a lot more data because of the number of vehicles theyve sold, says Miller. Problem is, it’s not clear what kind of data Tesla is collecting. Google’s cars capture the world with camera, radar, and laser sensors, with a human engineer along to flag the important stuff.
Mastery of self-driving under real-world conditions is not going to be easy.Raj Rajkumar, director of autonomous driving research at Carnegie-Mellon
Tesla’s cars are always online, but don’t think the automaker is building onall the data they provide. “It’s likely lots of data is thrown away, on the fly, while the car is driving,” saysJianxiong Xiao, formerly a computer engineer at Princeton University, now founder of a self-driving software startup.
Verdict: Tesla has more, and more diverse data, which might just be enough to leapfrog its competitors.
For extra difficulty,Musk is eschewing widely used LIDAR sensor technology, which fires out laser pulses to build a super accurate graphic representation of the environment.Musk has always hated it, and its price: What Google engineers call the “KFC bucket” on top of their car costs roughly $80,000.
Tesla is equipping its cars with eight cameras and a forward-facing radar, which Musk says will provide just as good a view of the world.
LIDARis a great technology, but its got moving parts which make it prone to fail, and its very expensive, says Miller. Unlike radar, it has trouble in rain, snow, and fog. Can Tesla do without it? Absolutely.
Not everyone is so sure. “Im quite skeptical they can have a solution, which is perfectly safe, without LIDAR,” says Xiao. He says Tesla might build a car that’s safer than a human driver (Musk says twice as safe), but that LIDAR is the only solution for 100 percent safety.
To translate sensor data into driving commands, Tesla is using equipping its cars with Nvidia’s Drive PX 2, a supercomputer that uses deep learning to teach the car to handle itself. For the cameras, Tesla will run in-house software, since its previous supplier, Mobileye, dumped it for moving too fast.
Verdict: Tesla’s laser-free approach is unconventional, but it can work.
Teslas ordering website offers a caveat: “Please note that Self-Driving functionality is dependent upon extensive software validation and regulatory approval, which may vary widely by jurisdiction.” No kidding.
Federal and state regulators in the US are slowly developing regulations to govern autonomous technology, but they’re nowhere near ready to validate Tesla’s proposed Level 5 system.
Musk is deliberately operating in a grey area, says Miller. “I think hes taking a big risk, but hes pushing the technology, he says.
Tesla plans toupgrade its cars’ autonomous capabilities in a drip, drip, fashion. One update could add the ability to obey traffic signals. Another could allow an empty car to search for a parking space, in a private lot, at low speed. The question is, when does the car become “autonomous,” and how do current and future laws govern its behavior?
Musk prefers the ask for forgiveness, rather than permission, approach. “It is a way of changing facts on the ground, and thats what Tesla, and more disruptive companies like Uberare really good at,” says Bryant Walker Smith, an assistant professor at the University of South Carolina School of Law who studies self-driving vehicles.
Verdict: Muskdoesnt want to wait for legislation to get his robo-rides on the road, but hemay have to.
Musk has put himself forcefully on the record as pro-autonomy, and says he’s not waiting for the hand-wringers to catch up. Car crashes kill 1.2 million people a year, more than 30,000 of them in the US. “The foundation is laid for cars to be fully autonomous, at a safety level we believe to be at least twice that of a person, maybe better,” he says.
Some worry that eagerness could set the technology back. “A serious concern is that Tesla brashly and prematurely introduces technologies that are not ready for deployment yet, encounters crashes and fatalities, and triggers significant pushback from society at large and perhaps much stricter regulation,” says Carnegie Mellon’s Professor Raj Rajkumar.
Yet problems with Tesla’s Autopilot system—one death in Florida, possibly another in China—haven’t generated serious fallout. “Weget increasingly big crashes and nothing seems to change,” says Walker Smith. German regulators asked Tesla to drop the term “Autopilot,”saying it could over-promise capability. Tesla basically just said nein.
Tesla’s self-driving capable cars are already onsale, even if they can’t do what Musk promises just yet. Now, the market will show if people want them. Early adopters—Teslas current core buyers—likely will. But the target marketfor the cheaper Model 3 is the general public. Whether theyll want to trust a computer, and Elon Musk, to drive them around will be answered by the sales figures.
Verdict: Musk’s ambitions usually don’t pan out quite on time, but they haven’t failed him yet.
Read more: http://www.wired.com/