If there's a voice of reason in the autonomous vehicle industry, Bryan Salesky is it.
One might think the CEO of Argo AI would have every reason to hype AVs, such as those being developed and tested by his Pittsburgh company. This year, Argo scored a $2.6 billion investment from Volkswagen Group, on top of a $1 billion commitment from Ford.
But Salesky makes it clear that self-driving vehicles are unlikely to become common on U.S. roads anytime soon. He reiterated his point in a Medium blog this month, declaring that AV technology "is not a winner-take-all sprint to the finish line."
"The opportunity ahead is bigger than any of us can imagine," Salesky wrote. "Its future will arrive gradually, and safely, if we do it right."
In an email interview with Shift Editor Leslie J. Allen, Salesky talked about his views on AVs and how a series of competitions staged more than a decade ago by DARPA, the Defense Advanced Research Projects Agency, helped jump-start the industry. Salesky, 39, led software development for Carnegie Mellon University's winning team in 2007. Here are edited excerpts.
Q: What is the legacy of the DARPA challenges?
A: The DARPA challenges were definitely an important catalyst for the development of self-driving technology. DARPA showed the world what was possible with autonomy. The three races created a drumbeat of progress that made the world take note of what is possible.
DARPA is a visionary organization. It rightly recognized the incredible potential of this technology for both government and commercial use. The races are responsible for creating pools of funding that generated research and talent in this area. Because of this, the United States is leading the way globally in self-driving technology. So what DARPA did was give a push to the technology and help show that it was possible to automate driving tasks at a level not yet seen.
How can self-driving technology change the lives of everyday people?
The reason why Argo AI is developing this technology is that we see the benefits it can bring to society. It has the potential to allow for significant improvements to transportation in terms of safety, accessibility, affordability and convenience. Over time, we can improve the geographic scope of operation, and the benefits to society can be tremendous.
People who are unable to afford the ride-hailing options of today might be able to take a self-driving vehicle to their job. Routes that are underserved by public transportation can be served.
When and if Level 4 and Level 5 AVs become our everyday vehicles, what part of driving would you miss the least?
I like to drive. However, I don't know anyone who likes to spend time stuck in traffic. If cities choose to adopt this technology, when it's ready, then there are potential benefits to reducing traffic congestion as well as the opportunity to offer a service that allows users to relax and spend their time in a self-driving vehicle instead of sitting in traffic.
But people should understand that self-driving cars are not a replacement for personally driven vehicles. We see the preservation of the freedom to drive a car as fundamental. If you want to drive, go for it! But a self-driving car can provide another option for transportation.
It's interesting to raise this question but we are so far away from this happening. It's important we educate and explain to consumers what to expect.
As you outlined in a 2017 blog, many complexities stand in the way of large-scale deployment of AVs. Which is the most daunting?
We have made an exceptional amount of progress since 2017. There are still technological hurdles. But we have a clear pathway to initial deployment. For large-scale deployment, we are working on leveraging the research community to push the frontier of the next generation of the technology. Operating at night, in fog and falling snow — those are challenges that have yet to be solved.
That's why we formed the Carnegie Mellon University Argo AI Center for Autonomous Vehicle Research. To fund the center, we have pledged $15 million over five years to fund a team of five of their top faculty leaders and support graduate students conducting research in pursuit of their doctorates. CMU will push the envelope on the next generation of self-driving technology.
We announced Argoverse, which comprises several datasets that can be used by the academic community for research. This will take time.
Another daunting aspect that's not as often discussed is that we must ensure that society is willing to accept AVs. That's as important as the technology itself.
Argo AI fleets operate in several cities. Whose streets are the toughest for the cars to negotiate?
We are testing in a larger geographic urban area than any other AV company we are aware of. All of the cities we are testing in present difficult challenges, which is why we picked them. The driving behavior and culture of people in each city is unique.
Take Pittsburgh, for example. They have the "Pittsburgh Left" turn. This means that the first person who is turning left is often allowed to turn in front of oncoming traffic once the light changes. This means the software needs to change for different geographic locations based on driving culture. Miami is particularly difficult because of the way pedestrians and other motorists behave. They're not rule followers. This makes it a great learning environment. In Detroit, we have potholes and lane markings that are not very clear.
In Washington, D.C., the traffic lanes change hour-by-hour sometimes. All of these things must be taken into account and they are helping us develop robust systems.
Having a diverse geographic footprint for testing will put us in a much better place to deploy services in a broad geographic area.
What's the biggest myth about self-driving cars?
Self-driving cars are not widely available today in the marketplace. It's misleading and downright wrong to market driver-assist features as self-driving. Camera and radar technology alone are simply not robust enough to develop a safe self-driving system.
If the driver is still responsible, it's not a self driving car. If the driver is required to monitor it, it's not a self-driving car. If the vehicle asks you to take back control, it's not a self-driving car. This concerns me because it's impacting consumer expectations for self-driving cars. So we are working to educate people on what the technology can do.
Self-driving cars used to be the stuff of science fiction. Speaking of that, what TV show or sci-fi movie did the best job of envisioning self-driving cars? Conversely, who got it wrong?
There are many films that we can look at for inspiration of what to do or not to do. Certainly, 2001: A Space Odyssey from Stanley Kubrick shows what a worst-case scenario looks like for a robotic control. No one wants HAL controlling their car.
One of my personal favorites is "Knight Rider." KITT was an inspiration that helped people imagine what an AI-driven car might be like. As someone who grew up in the '80s, I definitely watched this show and was motivated by it.午夜剧院