Once a month in a dilapidated Oakland warehouse with graffiti-filled walls, potholes and no electricity, two tech CEOs and a bunch of technologists gather to race autonomous vehicles.
Big tech companies like Alphabet, Tesla, Uber and others are racing to bring self driving car technology to market. IHS Market predicts 21 million in global sales by 2035, more than tripling to 76 million vehicles sold through 2035.
But these hackers and racers are strictly hobbyists.
"The funny thing is half these guys actually work for Google, they just don't work for the autonomous car division," said DIY Robocars group organizer Chris Anderson, who is also the chief executive officer of drone software company 3DR, formerly 3D Robotics.
One such Googler is Otavio Good, an engineer who helped create a Google app that lets users hold up a phone to some text in a foreign language -- like on a sign in a foreign country -- and see an instant translation.
"He's got great computer vision skills, great artificial intelligence skills, but doesn't happen to be on the autonomous car team so this is what he does for fun," said Anderson. (Google recently spun off its autonomous car unit into a separate company called Waymo held by parent company Alphabet.)
Though the technology hacked together by these weekend enthusiasts is not as sophisticated as the systems big tech companies are creating, Anderson is excited by how fast it is improving.
"It's getting better faster than Google did," he said.
Smaller vehicles, similar technology
Some members also race vehicles in a different event at an outdoor track at Thunderhill Raceway Park in Northern California.
At these events, teams kick off the day by training the vehicles' neural networks. Then come races — just like Formula One — in the afternoon. The end of the day is reserved for "wheel-to-wheel" racing, which involves a lot of crashing and is more like a Demolition Derby than a Formula One race, he said.
Most of the vehicles they race are on the small side — as small as a tenth the size of a regular car — making them cheap and disposable, said Anderson. Most are built at a cost of around $100, though there are some full size car projects that cost up to $15,000.
But even the tiny cars use software and sensors that is similar to the technology used by big tech companies and well-funded startups road-testing full-sized cars, he said.
Instead of doing the processing on-board, these robot cars tend to transmit the data from their sensors — cameras, sonar, lidar, radar and GPS, for instance — via Wifi to a laptop. From there, the engineers run professional-grade artificial intelligence and robotics software which controls the vehicles. All the code is open source and made available to the community.
Amateur enthusiasts were responsible for innovation in drones, something aerospace companies — often big government contractors — were not really interested in pursuing, he said. Why shouldn't the same be true with autonomous vehicles?
"We're competing with Google and Tesla — and they don't suck — so what can we possibly do as amateurs that they can't do?" he said.
"We're willing to take risks that they're not willing to take and we do this largely because we don't do full size cars," said Anderson. "Because we have lowered the consequence of failure…we have also increased the pace of innovation."
For his own part, Anderson partnered with Autodesk CEO Carl Bass to build and race a self-driving go-kart.
"This is how both of us get our kicks on weekends," he said. "He built the electric go-kart with his sons — because he's into manufacturing — and then I did the autonomy bit."
Being a human crash test dummy
Autodesk's Bass was also the "crash test dummy" at one Thunderhill race on April 1, said Anderson. "All he has is a red button — the steering wheel doesn't work — and the only reason he's there is we're still at the point where you have to press the button."
"Even that feels a little too scary to me," said Anderson, who prefers not to be in the hot seat.
At Thunderhill, teams tested two technological approaches: Systems based on so-called neural networks modeled after the human brain and those based on computer vision. No one has ever compared different types of autonomous driving technology in head-to-head racing before, said Anderson.
"The computer vision started faster but the pace of improvement was slower, whereas the neural networks started dumb but got smarter faster," he said. "Over the course of a weekend, we started to approach the best human times."
"By the end of the summer we'll have beaten humans, and then what?" he added.
There's also some friendly rivalry with sister organization the Self Racing Cars league, which was also there at the racetrack April 1, but racing full sized vehicles on a two mile track.
"These are the hottest startups in America and not a single one finished autonomously," he said. (Some did finish using GPS, which is sort of like cheating, he said.) "Meanwhile, we're right next door and we had 10 teams finish autonomously."
— This story has been updated to clarify the IHS Market prediction for autonomous vehicle market.