Introducing and a New Vision for Self-Driving

Today, we’re officially launching our company: We’re building artificial intelligence for self-driving vehicles. We think we have the technology to power this new era, and an important vision to make cars that are safe, trustworthy, and even fun.

But first, a bit on what drives us.

Every day, drivers and pedestrians are subject to a persisting danger: human error. Nearly 1.3 million people worldwide die in vehicle accidents every year, 90% of which are caused by human error. Over 33,000 of these deaths are in the US alone. That’s equivalent to a large passenger airplane crashing every single day. And someday, not too long from now, those deaths could be preventable. We started because we believe there’s a real opportunity to make our roads, our commutes, and our families safer.

We’ve now reached a point where every machine that can get smarter — from your phone to your refrigerator — will get smarter. Cars are at the forefront of this tectonic shift. We think that making them smarter and thereby safer is not a novel new feature, but an urgent mandate. And recently, an incredible new form of artificial intelligence has been moving through the technology industry, surpassing every benchmark and outperforming every metric. We think this new technology is the key to a future of self-driving vehicles. And today, we’re here to share our vision of this future, where transportation is safer, friendlier, and more reliable.

A powerful new form of AI

More than three years ago, six of our eight founders were graduate students studying at Stanford’s Artificial Intelligence Lab. They were looking at how deep learning — known as the most effective form of artificial intelligence — could be applied to the problem of self-driving cars.

Deep learning emerged onto the tech scene just a few years ago, when Google and other companies already had robust autonomous vehicle projects. Unlike other forms of AI, which involve programming many sets of rules, a deep learning algorithm learns more like a human brain. You provide examples, tagged and labeled by an expert, and the system starts to learn for itself — creating its own rules.

Say you’re driving down the road, and you see a bicyclist ahead of you. You know it’s a bicyclist — not because you count two wheels, and identify all the spokes, and see the handlebars. You know it’s a bicyclist because you’ve seen hundreds of bicyclists before, and you just know what they look like. That’s how our brains work, and it’s why we can drive cars and recognize faces, tasks that the many powerful rule-based algorithms struggle with.

The situations a car faces on the road are almost infinite. As we’ve been out testing our vehicles, we’ve seen people doing cartwheels and running around the car in circles. One time we saw a dog on a skateboard, wearing a helmet, trailing a person on roller blades. Using a traditional, rule-based approach, you’d have to program thousands of scenarios like these into your algorithm; miss one, and that could be the one that causes an accident.

But deep learning is different, and that’s what makes it so powerful. It has shown tremendous success understanding nuanced, variable situations. That’s why voice recognition products like Siri and Alexa use deep learning. Google uses it for image and web search, to understand the complexities of pictures and identify specific objects and people. And now, it’s time to apply it to driving.


At, we’re building the brain of the self-driving car. And in doing so, we aim to fundamentally reimagine the relationships between people, cars, and the world around them.

Our founders put their PhD studies on hold and founded for a simple reason. They knew that deep learning was the right technology to enable self-driving. This isn’t easy stuff; deep learning comprises some of the most complex, challenging algorithms in the engineering world.

We are at the forefront of the self-driving technology field because of years of experience developing deep learning software. We’re pushing this technology forward — from perception to decision making. Our goal is to empower the car to understand the world holistically, make the best decisions in a given situation, and communicate that decision to the driver and people outside the car. Moreover, we see a better, more accessible approach by taking full advantage of more cost-effective sensors, like cameras. From there, we take on perception, motion planning and controls in a smarter, more efficient way. Our approach to labeling and annotating data makes the learning semi-autonomous, removing the need for impractical manual tagging.

We think this approach fundamentally sets us apart from the rest of the industry. We’re not building the self-driving car piece by piece; instead, we wanted to reimagine the solution from the ground up.

The next era of transportation

Our company is built on the belief that deep learning is the safest, most effective way to teach cars to navigate the world. But getting from point A to point B is only part of the battle. Self-driving vehicles will maneuver through a world of nuance, inconsistency, and unpredictability. In other words: the human world.

Think about the four way intersection, when two cars arrive at the same time. Or the family that is hesitant to cross the street in front of a car at a stop sign. Or the highway lane that ends, forcing you to merge in front of someone. All of these situations require human-to-human communication. And there are so many more: bikers, parking garages, emergency stops, kids in the street, careless pedestrians, distracted drivers…

We humans have a language to navigate these situations: nods, hand waves, eye contact, polite honks… less polite honks. You simply can’t drive safely in the world without it. But with self-driving cars, everything changes. That’s why we’re focused on building a totally new language: one of trust and transparency. We shouldn’t fear self-driving cars. We should understand them, and feel safe around them, knowing exactly what the car is planning to do.

Self-driving cars need to build this trust through a user experience that is responsive and clearly communicates the car’s intent. Switching lanes, pulling over, and allowing pedestrians to cross all need to be signaled to riders, other drivers, and pedestrians. You design this experience with deep thought and intention from the start, and then put it out into the world. You can’t slap it on as an afterthought.

At, human-centered design is baked into everything we do. We don’t want people to just tolerate self-driving vehicles. We want people to love self-driving vehicles. Self driving isn’t just a new feature — it’s a once in a generation opportunity to reimagine the fundamental relationships between people, cars, and the world around them. On the way, we can make our roads safer, give people back countless hours of time, empower the disabled and the elderly, transform our urban landscapes, and cut down on CO2 emissions.

And just maybe, make cars fun again. That’s what we believe, and today we invite you to come along for the ride.

twitter: driveai_