Comma.ai: The future of autonomous cars?

Autonomous cars are the future. At some point you will no longer need to stay fixed to the steering wheel, staring intently into the distance as the mile markers roll by on the motorway for four hours. If you are lucky enough to have certain cars made by Tesla or Mercedes you might already be accustomed to flicking a switch and relaxing a little as the car takes control for a lot of the drive, however these systems are only Grade of Automation level 2, which requires that the driver pay attention at all times, and in fact modern Teslas will instruct you to keep your hands on the wheel, lest the autonomous system disables. That is unless you are clever/stupid enough to jam an orange in the correct spot to defeat the system (You can see it being done on youtube, but we do not condone it as it’s illegal almost everywhere and is just generally a bad idea).

All the large auto manufacturers are currently working on autonomous car technology, and they are all using a logic-based approach, thinking of the perfect way for a car to drive and then writing code to make the car do as it’s told based on the situation surrounding it. There is a small but disruptive startup that disagrees with this approach, however. Comma.ai, founded by hacker George Hotz, is using a different way to get cars to drive themselves, and there are benefits to his approach.

George Hotz became a big name when he was 17, and was the first person to carrier unlock the iPhone in August 2007. ]He traded his second unlocked iPhone for a Nissan 350z and 3 locked iPhones. Since then he has been employed by Facebook and Google and has continued to release security exploits for Android phones and iPhones allowing Jailbreaking or rooting.

As of September 2015 Hotz has been running his own startup called Comma.ai

The theory behind Comma.ai’s approach to autonomous driving emphasises learning from humans. To that extent, Comma.ai have released hardware that connects to your car’s internal computer, alongside a specialised dash cam, that records multiple parameters about your driving and sends it to Comma.ai’s servers. These servers use the collected data to build a driving model based on how humans drive, and in November 2016 they released the software that allows you to convert all the hardware used for data collection into hardware used for full self-driving. The total cost is less than $1000 if you have a car capable of running the hardware, at this point restricted to modern Toyotas, Hondas, Teslas and the Chevrolet itsolt. The Comma hardware piggybacks the driver assist devices already used by the car, the electronic throttle, ABS system and electronic steering to drive the car, with the dash cam acting as the “brain” of the whole system. All the while the car is operating in autonomous mode the dash cam continues to collect data allowing Comma.ai to refine their model, so the more people that use the system, the better the system becomes.

Being such a small startup Comma.ai don’t have the resources to develop the software allowing their hardware to talk to cars, so they also leverage their community for this too, providing monetary rewards for “porting” the Comma.ai driving model to other vehicles and for adding extra features. A large proportion of European cars use a different method for internal communications between the car’s computers compared to the typical and ubiquitous CAN bus used by their Japanese and American competitors, there is currently a $2000 bounty for the first person to allow a usb device to communicate with this different system, as it would allow for the Comma.ai ai software to be utilised by these European cars. 

At this point the best car for open pilot is the £25000 2017 Honda civic, far more reasonably priced than any other car with autonomous capability, but you do need to select a specific trim for the vehicle, usually a big expensive package of driver assist technology that on its own is useful, but combined with the Comma.ai hardware allows for very advanced mapping of the car and its place on the road. Combined with their carefully honed driving model, this allows improvements in the autonomy to be deployed with only simple software updates, and with an ever-growing community of people testing the self-driving, improving the driving model and porting the software to new vehicles, it seems like it will only get better