Have a spin and a chat with Comma.ai founder George Hotz in his $1,000 autonomous car

While on the road to develop plug-and-play autonomous driving technology, startup company Comma.ai needs data. Tons and tons of data.

Emme Hall Former editor for CNET Cars
I love two-seater, RWD convertibles and own a 2004 Mazdaspeed Miata for pavement fun and a lifted 2001 Miata for pre-running. I race air-cooled Volkswagens in desert races like the Mint 400 and the Baja 1000. I have won the Rebelle Rally, seven-day navigational challenge, twice and I am the only driver to compete in an EV, the Rivian R1T.
Emme Hall
4 min read
Watch this: We go riding with George Hotz and his $1,000 autonomous car

George Hotz, the whiz kid who was the first to jailbreak an iPhone back in 2007, has a new passion: AI. You may have read about his new company, Comma.ai, and his plans to beat Tesla, Google and Apple to market with self-driving technology that you can add to the car you already own. Now he's one step closer to his goal with a data-collecting app called Chffr, a riff on the word chauffeur.

George Hotz (Geohot)

Genius under 30? Possibly.

Bloomberg screengrab via Andrew Krok/Roadshow

Hotz's home-brewed AI technology uses human driver data to learn to autonomously pilot his company's 2015 Acura ILX test car on the highway. Comma.ai's goal of delivering aftermarket autonomous drive technology by the end of 2016 depends on having driver data. A lot of it. What's the easiest way to do that? Crowdsource it, naturally.

Chffr is like a cross between Dropcam and Fitbit. Open the app and mount your camera on your windshield, and it will record all your driving data -- things like how you react to other cars and bicyclists, your average speed and your braking and acceration patterns. It then uploads that data via Wi-Fi to Comma.ai's servers, where your driving behavior is analyzed alongside other drivers' datasets in order to help the company's autonomous driving technology learn. As with a Fitbit, you can dive deep into your own data, in this case to see where you drive, how far you drive and how long your commute takes.

To give people incentive to use the program, drivers earn "Comma Points" for each minute out on the road with the app activated. Hotz was quite cagey when asked what these points could eventually be redeemed for, saying only, "Comma Points are absolutely incredible and you'll wish you had them. You definitely want comma points. In a couple of months you'll be so happy you have Comma Points."

You can sign up to be a beta tester by going to Comma.ai and clicking on Chffr beta signup. Otherwise, Hotz says the app should be ready for Android, specifically the Samsung Galaxy S6 and S7 and the Nexus 6P, as well as Apple iPhone, by the end of June.


The final product will doubtlessly use something better than duct tape to attach the radar.

Emme Hall/Roadshow

We got to take a quick spin in Hotz's self-driving Acura prototype, in Las Vegas, Nevada. The technology uses a forward-facing radar, which for our demo was duct-taped to the nose of the car, and a forward-facing wide-angle camera mounted inside at the top center of the windshield.

It's worth noting that the test car made some more aggressive choices than the Delphi self-driving Audi I rode in a few months ago, but it still remained a safe distance behind the lead car. It still seems to have some trouble recognizing a stopped vehicle from a long distance away, forcing Hotz to intervene with manual braking, but did just fine approaching and stopping behind a stopped car at slower speeds.

But does the system learn? Can it adapt to a completely new situation? Hotz says he found out the answer a bit sooner than he anticipated upon arriving in Las Vegas.

Most streets in Vegas use Botts' dots as lane markers, those round, raised buttons that make a thump-thump sound under your wheel as you change lanes. The Acura had only driven on streets with traditional painted lane markers, but if Hotz is to be believed, his car was able to adapt to the new sensory input easily, and in short order.


The green markings are what the computer thinks are lane markings, the purple is the car's intended path. The end product may have a screen, but nothing near this size. Sorry, folks.

Emme Hall/Roadshow

The prototype system isn't quite perfect yet. Currently, the technology can't change lanes or recognize red lights, making it more useful for long drives on the highway and stop-and-go traffic.

Further, it apparently is so bad at reading Botts' dots in the twilight that Hotz insisted we take our test drive in full dark. However, Hotz is confident that the issue is the camera, not the technology, and that data collected via Chffr will help mitigate the problem.

Hotz claims he will have the technology ready to ship by the end of 2016 for $1,000. No word on the design of the product, but he claims it will be "as easy to set up as a piece of IKEA furniture" on your own car.

At the very minimum, your vehicle will need electronic power steering and an electronic stability control system in order for the technology to work. You do not need to have adaptive cruise control or lane-keeping assist already installed.

Comma.ai plans to have an update for city driving by the end of 2017. Hotz says his company is working with electric car manufacturer Wheego to install a surround camera and radar system to start gathering data for city driving.

Whether this technology will be ready for market in a mere seven months and, just as importantly, whether it will really work reliably remains to be seen. And, there's still the question of what drivers are going to do with all those Comma Points, which, for now, sound like just another nebulous metric in our increasingly gamified modern lives. But, the potential here is huge, and we're eager to see the final version of Hotz's hardware. Or, at least, one that doesn't require duct tape for installation.