It’s a damp evening in early January, and I’m riding shotgun in a Lincoln MKZ hybrid as it traverses the dark streets of Sunnyvale, California. Traffic is light, even though it’s rush hour on a Thursday, so we take twists and turns at a good clip. It would be a largely unremarkable ride, if not for the fact that the guy in the driver’s seat doesn’t have his hands on the wheel.
Traveling in Baidu’s self-driving car is comfortingly boring. The car, which ran the newest version of Baidu’s Apollo self-driving software, zipped along as the speed limit rose from 25 to 35 miles per hour, slowed down with balletic grace as we approached stop lights, and always—always—used its blinkers to signal turns.
Baidu, China’s giant search company, is a relative newcomer to the fast-growing autonomous-vehicle market, having begun work on its self-driving cars only about five years ago. Google started working on its self-driving-car project (now known as its Waymo business) back in 2009, and since then a number of tech companies and car makers have also invested heavily in the technology.
In an effort to catch up quickly, and raise China’s profile as an AI innovation center, Baidu is eschewing the secrecy that normally surrounds self-driving cars: as Google did with its Android smartphone operating system, it’s offering Apollo free to anyone who wants to use it.
Apollo 1.0 was released in July, and Baidu started testing Apollo-running cars on public roads in late 2017. Baidu hopes companies that use Apollo—it has 90 partners so far, including car makers like Lincoln owner Ford, car-component makers like Continental, and chip makers like Nvidia—will then contribute data that it can use.
Over time, this will make the software powering rides like the one I took better, faster, and safer. In many ways, the approach is like the one Google used to help Android become the world’s most popular OS.
Jingao Wang, senior director of Baidu’s intelligent driving group and the head of Apollo, spent years working on Android early on, helping with the release of the first several versions of the OS and early Android smartphones. With autonomous driving, he says, the open-source approach makes a lot of sense simply because it’s an AI-based technology that needs a huge amount of data to thrive.
Self-driving cars for private transportation aren’t ready yet. For instance, I couldn’t just jump in that sensor-laden car and command it to take me anywhere I wanted. Although the loop it drove was not extensively tested in advance, it was driven several times over two or three days to build a high-definition map that could take into account things like which traffic lights match up with which lanes of traffic.
That said, buses and shuttles all over the world drive in prescribed loops, and Wang says the company plans to try out a self-driving mini-bus in China this year in a limited area like an industrial park (Google’s Waymo is already testing a self-driving taxi service with some consumers). Baidu is also launching a test in Los Angeles of short-distance autonomous-vehicle rides for the disabled; it’s slated to begin by the end of the year. Baidu says it will work with a Chinese bus maker to manufacture driverless buses by the end of 2018, too.
“Autonomous driving is a once-in-a-century technology to make the world better,” Wang says.
Despite this optimism, Baidu, like other companies chasing autonomous dreams, is facing some very real obstacles. There’s still work to be done in perception, as Wang acknowledges, both in understanding what’s around the vehicle—cars, cyclists, pedestrians—and in predicting what may happen next.
I experienced this firsthand during my ride. At one point, the car suddenly slowed down in anticipation that a driver might pull out in front of it (which didn’t happen).
One obvious challenge: when multiple people using different forms of transportation approach a four-way stop at the same time. Wang says Baidu’s self-driving vehicles will respond “on the polite side” when encountering a human at a stop sign and act cautiously around cyclists to calculate enough space to make room for them on the road; beyond that, he suggests, the car may somehow signal that someone else should go first at an intersection.
“There’s still a ways to go,” he says. “When we’re driving around Sunnyvale, we’re not talking about San Francisco, or even Beijing.”
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
The Biggest Questions: What is death?
New neuroscience is challenging our understanding of the dying process—bringing opportunities for the living.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
How to fix the internet
If we want online discourse to improve, we need to move beyond the big platforms.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.