Among the teams building autonomous vehicles, few things are debated as passionately as the best sensor combination. The gist is this: picking wisely between LIDARs, cameras, and radar could be the difference between getting to market at a viable price point or being run over by your competition. Alphabet’s Waymo, Uber, Ford, and GM have all placed big bets on LIDAR (Light Detection and Ranging) sensors. But others, including startup Comma.ai and Tesla, argue that you don’t need LIDAR at all.
To get an expert’s perspective, we talked to our friend Angus Pacala
, co-founder of the LIDAR company Ouster
. Ouster is Angus’s second LIDAR company, so he’s no stranger to the debate.
Angus, can you weigh in on this ongoing debate between LIDAR proponents and those who claim that radar and vision technologies are sufficient?
I think it’s a misplaced debate driven by the current price point of LIDAR. If you’re promising affordable Level 4 self-driving and LIDAR is prohibitively expensive, you have to find a solution with vision and maybe radar. In the next five years though, multi-hundred-dollar automotive LIDAR sensors will exist. There won’t be any reason not to use LIDAR. The more interesting question is how all these sensor modalities are going to come together to achieve much higher performance.
You see a mix of sensors being used by the airline industry without any debate over which are necessary or extraneous. You can fly a Cessna with just your eyes, but we would never talk about removing radar altimeters and autopilots from airplanes. They’re economical ways to enhance safety, so no one questions the need for them. When LIDAR’s cheap enough, this debate is going to fade.
So how do you bring down the cost of LIDAR down to a point where this isn’t a debate?
Well, that’s the crux of what we’re working on at Ouster. We’re doing this by leveraging the semi-conductor industry, or as we say by bringing Moore’s law to LIDAR.
With computer chips, we’ve seen these massive 2x or 3x density improvements every 24 months for 50 years, and there’s been a similar trend for the technology behind CMOS digital cameras and other silicon photonics. The camera in a modern smartphone is higher resolution and has better low-light performance by a wide margin than what was possible 10 years ago. It’s an incredible achievement at many levels – the physics of the photosensors themselves, the performance of CMOS analog and digital circuitry, and the improvement in miniature optics manufacturing.
We’ve decided to leverage all that progress by replacing thousands of legacy components with a single custom semiconductor chip that senses light and does massive amounts of processing too. It’s the only route to achieving the relentless improvements in both performance and cost that the industry needs.
Do you think a new price point in LIDAR will unlock new markets and applications?
Absolutely. In ten years, I think that improvements in semiconductors will support very inexpensive solid-state LIDAR. That’s going to unlock many new consumer and industrial applications. Even today, you wouldn’t believe the diversity of companies that reach out to us looking for high-quality 3D sensors. You wouldn’t believe the bizarre things that people want to survey; it turns out people want full-on 3D models of a lot of different structures.
I think we’ll also see a bifurcated LIDAR market. While many new applications will be unlocked by a lower price point, plenty of applications will continue to place performance at a premium and support higher price points.
So if you were sitting at a bar with the premium LIDAR players in the market – say, Velodyne and Luminar – what would you disagree on?
I think we’ve taken a very different approach. We disagree with folks who say LIDAR can’t be a competitively priced sensor. Velodyne and Luminar are super premium products, but our belief is that performance at all costs means that you can’t get the technology into the hands of enough people.
Also, this is not so much of a disagreement, as it is a difference in go-to-market. There’s a lot of noise and hype coming from the 50-plus companies building LIDARs. It’s pretty easy to show rendered simulations or cleaned, filtered data that overpromise and under-deliver. From the start, we’ve shown our customers totally unprocessed data that stands on its own merits. We took our time in stealth and now we’re confident we have something real – something that our customers need and want. Our rallying cry has always been, “Ship products, not promises.”
Thanks, Angus, for taking the time to share your thoughts with us! You can see videos of Ouster’s OS-1 here and here. If you have thoughts on the LIDAR and sensor-software debate, we invite you to share them with us by sending us an email at: firstname.lastname@example.org