Decentralizing the AV Market with Comma.ai: Part 1

When you think about self-driving cars, what company comes to your mind first?

Is it Tesla, Google, or even BMW?

Typically, other smaller players fall by the wayside in the face of such competition. To make matters worse, the general consensus seems to be that self-driving cars will not be a viable, affordable option any time soon.

What might help to change all of this are the efforts of a company like Comma.ai.

Before we jump into what might make Comma a unique contender in the AV niche, it’s important that you get familiar with their unique founder, George Hotz. When he was only 17, he reportedly was the first person to ”jailbreak” the iPhone or release it for usage globally with any carrier. According to one of our sources below, he actually followed this up by hacking the PlayStation 3 to the extent that Sony sued him. In the end, the case ended in a settlement and Hotz went on to work at both Google and Facebook, which are opportunities he seems to have at least partially leveraged with his hacking fame. In 2015, however, he left tech royalty to work at a company called Vicarious, which to this day, is still determined to usher in the age of Artificial General Intelligence. In essence, Vicarious was inspired by theoretical concepts related to programming the human brain to put all of their resources into developing AI systems that can truly learn. In Hotz’s case, he had always been interested in such work but after reading the bulk of Vicarious’s research, he began to feel that something was missing.

After only about six months on the job, Hotz felt confident enough to quit and go directly to Elon Musk at Tesla to propose a new computer vision system for all of their existing vehicles. Because this sort of system might be difficult to conceptualize, just imagine a group of sensors and cameras that help a self-driving vehicle detect dangers on the road. In Hotz’s case, it is also safe to assume that his pitch included an embedded AI system as well. While in the end, this potential partnership did not really work out, it did catalyze the creation of Comma.ai.

Shortly after Hotz left Tesla with nothing to show for his efforts, he pitched the idea of creating a consumer-friendly self-driving vehicle system to the Andreessen Horowitz venture capital firm. Reportedly, they liked his idea enough to reward him with $3.1 million in seed capital. For the new two years, the Comma.ai team, which started with only four people, began to put nearly all of its’ money into product development and testing.

As of now, Comma has four products, which come together to form a sort of interactive ecosystem with the goal of democratizing access to AVs. Central to this ecosystem is Comma’s open-source software called openpilot, which functions as a driver assistant in a few key ways. What these are will become clearer once you understand what its’ complementary products are meant to bring to the table.

In this particular piece, we’ll mention two of these three gadgets with the aim of saving the final discussion for our next post. First and foremost there’s Eon, a souped-up dashcam, that records all kinds of training data for Comma’s principal AI system. In essence, the aim here is to use supervised learning principles to drive continuous improvement in their AI and therefore also in their overall platform. At this juncture, it already seems clear that the more users Comma is able to acquire and retain, the more accurate its’ software will be. For this to happen, however, all of the many moving pieces of the Comma ecosystem will have to work well as one. With the overall complexity of Comma’s platform in mind, we’ll stick to one more product for now, which is called Panda, along with its’ complementary parts.

In the most basic sense, Panda is a small dongle that can connect your car to your phone or your computer over WiFi. The point of doing this is actually to allow you to ”speak” directly to your car, in the sense that you will have the power to change things about it that the average car owner would not have. If this still isn’t clear to you then rest assured, you are far from alone in feeling that way.

It makes much less sense alone than it does within its’ own mini-platform, which includes an app called Chiffr and a set of software called Cabana. When all of these pieces work together, they essentially function like a singular AI system that takes in your car’s data, analyzes it, and then makes conclusions on it that Comma’s ideal customer can understand. Imagine having access to a continuous data feed of real-time statistics on everything about your car.

From here, arguably, things get even more interesting. These three products are not meant for the average person. They’re basically akin to a software development kit for hackers and people who are very knowledgeable about the inner workings of a car already. With Panda, Cabana, and Chiffr, Hotz and his team hope that they can incentivize these groups to help them continuously add to their openpilot software. In our next posts, we’ll take a deep dive into open pilot as well as the remaining pieces that allow it to function as it should.

Until then, keep in mind that through crowdsourcing training data, Comma.ai hopes to vault over their richer counterparts who keep all of their research and development in-house. If this strategy works, it could upend the way the automotive industry does business even beyond the AV niche.