Render of autonomous self driving driverless vehicle finding parking on road by using lidar
An arrow pointing leftHome

Lumotive’s solid-state lidar partner program is good news for self-driving car safety

  • Mike Pearl

No more spinning coffee cans on roofs could mean more sensitive AI in cars.

If I ran up to you and said “We can all breathe a sigh of relief now that the chips from Lumotive’s Meta-Lidar solid-state lidar systems are starting to be provided to other companies for use in their AI systems!” you would … probably not feel all that relieved. This news from Seattle-based lidar developer Lumotive would probably have you asking yourself: “Alright … so what does this mean for me?” So let me explain:

In May of 2016, a Tesla Model S driver in Florida became the first known death behind the wheel of a self-driving car. It couldn’t have helped that the driver was likely watching a Harry Potter movie at the time of the crash, but investigators also noted that the camera-based imaging system in the Tesla was unable to distinguish between the glaring bright sky, and the glaring white side of the tractor trailer the car slammed into.

It was the sort of optical illusion that can trick a human eye, or a camera, but it would never trick those clunky spinning coffee can-like things you sometimes see on top of self-driving cars: lidar.

Lidar, which is short for ‘light detection and ranging’, aren’t cameras. They’re essentially lasers that scan their surroundings and make 3D images, creating a collection of points in space that look less like a photo, and more like the way The Terminator sees the world. In some ways, they’re inferior to a camera. Rain can confuse lidar, for instance. But shiny semi-truck trailers can confuse cameras, with tragic results. So lidar can be one of multiple redundant systems feeding information into a self-driving AI.

But as you’ve probably noticed, Teslas don’t have clunky spinning coffee cans on them. Self-driving cars used by Alphabet’s Waymo have them mounted on their rooftops, and you can’t miss them. But Tesla CEO Elon Musk has called lidar “lame,” and relies, instead, mainly on cameras, for his Autopilot product (which, contrary to the name, is not fully automated and still requires drivers to pay attention and not watch “Harry Potter” while driving). Tesla used to use radar in addition to cameras, but that stopped earlier this year.

Lumotive CEO Sam Heidari stated in a press release that, “Increasingly, 3D sensing application performance requirements can only be addressed with a solid-state design.” And Lumotive’s Light Control Metasurface beam steering chips — the chips in its Meta-Lidar product — are just that: lidar systems with no moving parts. And while we may still be in the early days of solid-state lidar, this is promising news for, among other things, automotive safety.

So, while typical lidar has to spin around to sweep an area with its laser over and over — requiring complicated and expensive sensors built into the car — Lumotive’s LCMs instead rely on subtle changes in the surface of the chip to steer the beam while remaining stationary. With no spinning coffee cans necessary, one can easily imagine a lidar system installed on a mainstream car.

Andrej Karpathy, Tesla’s head AI scientist, seems to be pretty committed to using cameras instead of lidar, though Tesla’s with lidar onboard have been spotted by fans. What’s more, Tesla is not the only player in the self-driving game. Cadillac’s Super Cruise driver assist system is widely regarded as better than Tesla’s, and it currently uses lidar, though the sensors are expensive and bulky. Wider availability of solid-state lidar makes it much easier to imagine a future in which automotive designers are free to use lidar in their self-driving systems.

But there’s still a ways to go before these will be ready for installation on cars, and Lumotive’s “Technology Access Program,” which was expanded earlier this month, makes that abundantly clear. In addition to the chips and transmitters, Lumotive’s partner program gives participants an “evaluation kit” for the solid-state chip, a “reference platform” for designers, an application programming interface for programmers to use, a cache of data about the technology and help from engineers and tech support personnel.

If automotive applications for solid-state lidar are perfected in the near future, maybe the elusive level-5 autonomy isn’t so far off. And if that’s true, maybe people like me — people who absolutely loathe driving — will live to see the fully hands-free, and fully safe, self-driving promised land. Or we can dream, anyway.