Here We Go, Frisco:
Our Self-Driving Service is Live, With One Million Simulated Miles Under Our Belt
Time to Ride
We announced the news of our self-driving service back in May, and now the time has arrived: today the pilot is launching in Frisco, Texas. At Drive.ai, our mission is to create self-driving technology that solves people’s transportation problems — today. That’s why we’ve been collaborating with Frisco TMA to design a program that serves the community and brings them to and from the places they want to go. Rather than creating tech for tech’s sake, we’re working to help people get where they need to go in a way that’s safe, efficient, and enjoyable to experience.
In April 2018, we touched down in Texas and began driving along the route in Frisco, collecting data to inform our deep learning AI systems, documenting the scenarios we encounter, and creating custom simulations to improve the way our vehicles perform on the roads. In the last four months, we’ve been driving the streets of our geo-fenced route in Frisco, which crosses six lanes of traffic, takes us through parking lots, and involves pulling over to pick up and drop off passengers on-demand. We’ve been focused on creating a great user experience for those who will be hailing and experiencing these rides, and have been working with the city to educate the community and hear what people have to say.
Our focus is on solving a mobility challenge, and doing so safely. To that end, in addition to our “real world” logged miles, we’ve racked up over one million simulated miles on our Frisco route. For us, the one million marker is more than a milestone; it’s a number that reinforces our ability to deliver a safe, thoroughly-tested service to the public.
Seeing It All With Simulation
Simulations are a major component of our approach to self-driving, as running comprehensive simulations improves our vehicles’ ability to foresee and handle a wider variety of driving scenarios — those that are common, and those that are less so. This is critical to ensuring the depth of our vehicles’ understanding of the world around them, and in turn, the safety of our passengers, other drivers, cyclists, and pedestrians who share the roads with us.
When we create simulations, we do so in two ways: by replicating scenarios we’ve encountered in the “real world,” and by creating our own.
As we examine our driving logs from autonomous rides on our routes, we can choose to convert the “real world data” into simulations that we want to examine more closely. It’s like a high tech version of SimCity, where we design the world, and can then replay events and modify their components to explore how our technology responds in unique scenarios. This is a good place to start for the more common things that people do on the roads: navigating tricky intersections, right-of-way decisions, and observing the behaviors of cyclists and pedestrians.
On the other hand, we can also create original simulations that introduce some of the more unusual situations that we want our vehicles to be prepared to handle. In this way, we can specifically craft unusual situations for our vehicles, like piloting around cars that are double parked, or carefully steering through tight turns. By deliberately exposing our technology to “stress tests” like these, we are able to ensure a level of competency in complicated circumstances we may not encounter organically.
The Human Factor
With our self-driving program officially launching in Frisco, safety and trust are our top concerns. Our vehicles drive like the most respectful, by-the-book drivers out there; and remarkably uneventful rides are exactly what we’re going for, every time. While our vehicles are designed to carefully heed all traffic rules, the fact of the matter is that we can’t control how other drivers behave.
So, how do we ensure that our vehicles still operate as safely as possible even when the world around them may not be as predictable? That’s where simulation steps up, helping us create scenarios that include ‘unlikely’ human behaviors — like people darting across the road or objects rolling into the road, driving through crowded or sparsely populated streets, navigating around poorly parked cars, or riding alongside a cyclist for an extended period of time. By introducing “dynamic agents” such as these, we have expanded the scope of scenarios our vehicles might experience and can confidently address.
Within these scenarios, we can then get even more specific: adjusting various parameters to new sizes and shapes, to see how our cars will react if things were slightly different. For example, we might change the angle of a double-parked car, or the size of a left-hand turning lane. We test our simulations to exhaustive detail, so that our vehicles know how to approach these variable driving conditions and situations, in every combination that could be possible.
One Million Down, Many More to Come
We’re launching our on-demand Frisco service to provide a convenient option for people looking to easily get from point A to point B, and we want our riders — and all others on the roads — to have the utmost confidence and comfort in our vehicles. That’s why we’ve prioritized our vehicles’ ability to operate safely in simulated settings, which improves their performance on the roads each day.
While we have logged one million simulated miles on the streets of Frisco, we’re not stopping there. With each mile driven — both on the streets and in simulations — we will continue to improve our technology. Today is the exciting start to our on-demand ride service, which we hope will benefit a community and increase understanding of and confidence in self-driving technology.