Lidar (Light Detecion And Ranging) is essential for self-driving cars — here's how some leading lidar sensors work.
By Timothy B. Lee, Feb 1, 2019 | Original ARS Technica article here.
Lidar, short for light radar, is a crucial enabling technology for self-driving cars. The sensors provide a three-dimensional point cloud of a car's surroundings, and the concept helped teams win the DARPA Urban Challenge back in 2007. Lidar systems have been standard on self-driving cars ever since.
In recent years, dozens of lidar startups have been created to challenge industry leader Velodyne. They've all made big promises about better prices and performance. At the start of 2018, Ars covered the major trends in the lidar industry and why experts expected cheaper, better systems to arrive in the next few years. But that piece didn't go into much detail about individual lidar companies—largely because most companies were closely guarding information about how their technology worked.
But over the last year, I've gotten a steady stream of pitches from lidar companies, and I've talked to as many of them as I could. Ars has now been in contact with senior executives from at least eight lidar companies as well as others involved in the industry as customers or analysts. These conversations have provided a lot of insight not only into trends in the lidar industry in general but also about the technology and business strategy of individual companies.
Today, there are three big ways that lidar products differ from one another. And after laying these approaches out, it's easier to grasp the technology of nine leading lidar companies.
To keep this survey of the lidar landscape manageable, I'm sticking to independent companies that focus primarily on the lidar business. That means I won't cover Waymo's homebrew lidar technology, the lidar startups GM and Ford acquired in 2017, or the lidar efforts of bigger companies like Valeo (maker of the lidar in Audi's 2018 and 2019 versions of the A7 and A8), Pioneer, or Continental. It's hard to get these larger companies to give us details about their lidar technology—and there's plenty of ground to cover without them.
The Three Big Factors That Distinguish Lidar Sensors
The basic idea of lidar is simple: a sensor sends out laser beams in various directions and waits for them to bounce back. Because light travels at a known speed, the round-trip time gives a precise estimate of the distance.
While the basic idea is simple, the details get complicated fast. Every lidar maker has to make three basic decisions: how to point the laser in different directions, how to measure the round-trip time, and what frequency of light to use. We'll look at each of these in turn.
Beam-steering technology: Most leading lidar sensors use one of four methods to direct laser beams in different directions (two companies I cover here—Baraja and Cepton—use other techniques that they haven't fully explained):
- Spinning lidar. Velodyne created the modern lidar industry around 2007 when it introduced a lidar unit that stacked 64 lasers in a vertical column and spun the whole thing around many times per second. Velodyne's high-end sensors still use this basic approach, and at least one competitor, Ouster, has followed suit. This approach has the advantage of 360-degree coverage, but critics question whether spinning lidar can be made cheap and reliable enough for mass-market use.
- Mechanical scanning lidar uses a mirror to redirect a single laser in different directions. Some lidar companies in this category use a technology called a micro-electro-mechanical system (MEMS) to drive the mirror.
- Optical phased array lidar uses a row of emitters that can change the direction of a laser beam by adjusting the relative phase of the signal from one transmitter to the next. We'll describe this technique in detail in the section on Quanergy.
- Flash lidar illuminates the entire field with a single flash. Current flash lidar technologies use a single wide-angle laser. This can make it difficult to reach long ranges since any given point gets only a small fraction of the source laser's light. At least one company (Ouster) is planning to eventually build multi-laser flash systems that have an array of thousands or millions of lasers—each pointed in a different direction.
Distance Measurement
Lidar measures how long light takes to travel to an object and bounce back. There are three basic ways to do this:
-
Time-of-flight lidar send out a short pulse and measures how long it takes to detect the return flash.
-
Frequency-modulated continuous-wave (FMCW) lidar sends out a continuous beam whose frequency changes steadily over time. The beam is split into two, with one half of the beam getting sent out in the world, then being reunited with the other half after it bounces back. Because the source beam has a steadily changing frequency, the difference in travel distance between the beams translates to slightly different beam frequencies. This produces an interference pattern with a beat frequency that is a function of the round-trip time (and therefore of the round-trip distance). This might seem like a needlessly complicated way to measure how far a laser beam travels, but it has a couple of big advantages. FMCW lidar is resistant to interference from other lidar units or from the Sun. FMCW lidar can also use Doppler shifts to measure the velocity of objects as well as their distance.
-
Amplitude-modulated continuous wave lidar can be seen as a compromise between the other two options. Like a basic time-of-flight system, AMCW lidars send out a signal and then measure how long it takes for that signal to bounce back. But whereas time-of-flight systems send out a single pulse, AMCW systems send out a more complex pattern (like a pseudo-random stream of digitally encoded one and zeros, for example). Supporters say this makes AMCW lidar more resistant to interference than simple time-of-flight systems.
Laser wavelength
The lidars featured in this article use one of three wavelengths: 850 nanometers, 905 nanometers, or 1550 nanometers.
This choice matters for two main reasons. One is eye safety. The fluid in the human eye is transparent to light at 850 and 905nm, allowing the light to reach the retina at the back of the eye. If the laser is too powerful, it can cause permanent eye damage.
On the other hand, the eye is opaque to 1550nm light, allowing 1550nm lidar to operate at much higher power levels without causing retina damage. Higher power levels can translate to longer a range.
So why doesn't everyone use 1550nm lasers for lidar? Detectors for 850 and 905nm light can be built using cheap, ubiquitous silicon technologies. Building a lidar based on 1550nm lasers, in contrast, requires the use of exotic, expensive materials like indium gallium arsenide.
And while 1550nm lasers can operate at higher power levels without a risk to human eyes, those higher-power levels can still cause other problems. At the CES show in Las Vegas this year, a man reported that a powerful 1550nm laser from an AEye lidar permanently damaged his camera. And, of course, higher-power lasers consume more energy, lowering a vehicle's range and energy efficiency.
With this background out of the way, let's look at 10 of the leading lidar companies.
Velodyne
Three Velodyne products: the Alpha Puck, Velarray, and Veladome.
Beam steering: Spinning
Distance measurement: Time-of-flight
Wavelength: 905nm
Velodyne invented modern three-dimensional lidar over a decade ago, and the company has dominated the lidar market ever since. Velodyne's distinctive spinning lidars continue to be ubiquitous on self-driving cars, and the company is likely to continue leading the market in 2019. But some industry observers question whether Velodyne can maintain its industry dominance in the coming years.
As recently as late 2017, Velodyne's flagship 64-laser lidar unit was selling for a reported $75,000 each. Velodyne introduced a new 128-laser model that's rumored to be even more expensive—as much as $100,000.
Asked about these figures, a Velodyne spokesman replied: We don't reveal prices in the public domain, but the prices that are quoted in public domain are list prices for single units. The prices are significantly lower in automotive volumes and we are actively delivering to automotive OEMs at these lower prices.
Velodyne does sell less-expensive lidars, including a 16-laser puck
model that was selling for $4,000 last year. Velodyne also has a solid-state model called the Velarray. Velodyne says that it's a 905nm system with a proprietary frictionless beam-steering method.
Velodyne expects it to eventually cost less than $1,000 in automotive volumes. However, these lidars do not deliver the high-end performance of Velodyne's spinning 64- and 128-laser models.
Some critics claim that Velodyne has struggled with manufacturing and product quality.
The delicate moving lidar sensors that are its bread and butter have proven difficult to manufacture efficiently at high quality and can be frustratingly fragile in automotive applications,
journalist Ed Niedermeyer wrote recently, citing sources in the autonomous vehicle sector.
A company spokesman disputed this characterization, saying that Velodyne has over the years perfected the science of manufacturing these sensors at scale
and has been proven to withstand harsh automotive grade environments.
Velodyne recently signed a licensing deal with Veoneer, an established company in the automotive supply chain. Veoneer has plenty of experience building components that meet car companies' exacting quality standards, and it may figure out ways to tweak Velodyne's classic design in ways that improve quality and bring down costs. But they'll have to move quickly, as a number of other companies are aiming to take Velodyne's lidar crown.
Luminar
Beam steering: Mechanical scanning
Distance measurement: Time-of-flight
Wavelength: 1550nm
Luminar is widely seen as one of Velodyne's leading rivals. The company has been in business since 2012, and it began volume production of its lidar units last year. And the company argues that it has industry-leading performance.
That's partly due to Luminar's decision to use 1550nm lasers. Using an inherently eye-safe wavelength allows Luminar to crank up the power on its lasers, which helps it to see farther. Using 1550nm lasers means Luminar has to use exotic indium gallium arsenide sensors to detect return flashes. That ought to be expensive, but Luminar told us last year that the receiver assembly in its lidar units costs only $3.
When we asked Velodyne president Marta Hall about Luminar last year, she pointed to Luminar's power consumption as a significant downside. That's particularly significant because Luminar's lidar are fixed units with a field of view of 120 degrees. That means you need four Luminar units (with some overlap) to provide the same 360 degree field of view as a lidar unit from Velodyne or Ouster. (Update: In an email statement, Luminar said that the latest version of its lidar has significantly lower power consumption than early models: around 50 watts all-in.
)
Luminar has also been tight-lipped about prices. Last May, Luminar CEO Austin Russell told us that lidar will need to get down to low single-digit thousands
to be viable in the consumer market and that this is not an issue
for Luminar. But that implies that the company's units were significantly more than the low-single-digit thousands at the time.
Luminar is ahead of many lidar makers when it comes to actually shipping products to customers, having begun volume production more than nine months ago. Over the last 18 months, Luminar has scored partnerships with Toyota, Volkswagen, and Volvo.
In a recent interview, Russell pointed to those deals as one of Luminar's biggest competitive advantages. Major companies are designing self-driving systems around Luminar's lidar, Russell told me, and that will make it costly for them to switch to a competing lidar provider in the future.
AEye
Beam steering: Mechanical scanning
Distance measurement: Time of flight
Wavelength: 1550nm
AEye has much in common with Luminar. It uses a mechanically scanned mirror for beam steering. It uses a laser at the eye-safe wavelength of 1550nm, allowing it to run at high power levels. As a result, AEye's lidar boasts impressive range figures. AEye says that its lidar can see as far as 1,000 meters away—vastly more than the 200 to 300 meters most high-end lidars claim.
In a December interview, AEye CEO Luis Dussan touted the high-power bursts that can be produced by the fiber lasers in AEye lidar. Plenty of other lidar products are based on diode lasers that are limited to 100-150 watts,
he said. Fiber lasers can go up to 100,000 watts—very short pulse, huge amount of signal.
High power levels enable long range, but they come with significant downsides. During this year's CES show in Las Vegas, a man told Ars Technica that his high-end camera had been permanently damaged when he snapped a photo of an AEye lidar demonstration. Eyes are filled with fluid that's opaque to 1550nm light. Cameras aren't. And so the high-powered AEye laser apparently became concentrated on the fragile image sensor in the man's camera.
In a statement to Ars, AEye described camera damage as an industry-wide problem. But Angus Pacala, CEO of rival Ouster, disputed that. He wrote Our sensors are camera and eye safe. Period.
And Luminar wrote that we’ve conducted extensive testing on the same camera with the same lens and same settings that was damaged at CES
and were no able to cause damage with Luminar's lidar.
Most lidars use a fixed scanning pattern. AEye's lidar takes a different approach that the company calls agile scanning.
The scanning pattern of AEye lidar can be configured in software and adjusted dynamically. According to Dussan, AEye's agile scanning pattern worked together with the flexibility of its fiber laser. We can control the pulse energy from shot to shot,
he told Ars. Software controls not only where the next measurement occurs, but also how much power is used—and therefore the range—in the next measurement.
Hence, if the lidar spots a faraway object, it can pump up both the scanning resolution and the power level in that portion of the image, providing many more data points. That can yield a high-resolution scan that can help to distinguish between a pedestrian, a motorcycle, or a large piece of trash left on the side of the road.
On the other hand, there's a danger of over-optimizing. If a lidar system spends a lot of time scanning objects it has already recognized, it creates a greater danger that it could devote too little time to systematic sweeps, missing other objects as a result.
Ouster
Beam steering: Spinning
Distance measurement: Time-of-flight
Wavelength: 850nm
At first glance, Ouster's lidar looks a lot like Velodyne's lidar. Both are mechanically spinning time-of-flight systems, and both companies sell units with 16, 64, and 128 lasers. That's not a coincidence: Ouster deliberately designed its products to be drop-in replacements for Velodyne's units, because many potential customers have grown comfortable with the classic Velodyne form factor.
But if you crack open Ouster's lidar units, they look very different on the inside. Velodyne's classic design used 64 individually packaged lasers and 64 individually packaged detectors, according to patent filings. In contrast, Ouster has figured out how to pack 64 lasers onto a single chip, with a second chip containing 64 sensors to detect the light that bounces back. This integrated design has the potential to dramatically reduce the cost and complexity of manufacturing lidar units.
Ouster's most sophisticated lidar, due to ship later this year, is the OS-2, a 64-laser unit that retails for $24,000. Ouster says this unit offers range comparable to Velodyne's high-end 64-laser unit. Ouster also sells shorter-range units for as little as $3,500.
Ouster is able to pack 64 lasers onto a chip using a technology called a vertical-cavity surface-emitting laser—in contrast to the edge-emitting lasers used by so many other lidar makers. Because VCSELs emit light perpendicular to the surface of the wafer, it's possible to pack many lasers onto a semiconductor die. The technology has long been used for consumer applications, like computer mice, but has traditionally not been considered powerful enough for lidar use. Ouster says it's figured out how to build a high-performance lidar unit using VCSELs.
Ouster uses another semiconductor technology called single-photon avalanche diodes for detecting returned light. As with VCSELs, SPADs can be manufactured using standard silicon-based chip techniques, and many SPADs can be packed onto a single die. That made it relatively easy for Ouster to upgrade from the 64-laser units it was selling last year to a 128-laser unit it announced in January and plans to start shipping this summer. The company just had to replace the 64-laser and 64-detector chips it used in the old model with 128-unit versions.
Upgrading from 64 lasers to 128 lasers is just the start, argues Ouster CEO Angus Pacala. Within a few years, Pacala expects to introduce lidar units that have thousands—and perhaps eventually millions—of VCSEL lasers and SPAD detectors.
So far, Ouster has focused on building one-dimensional arrays of lasers for use in a Velodyne-style spinning sensor. But Pacala says the same technique can be used to build two-dimensional arrays of lasers and detectors—much like the image sensor in a camera. That could enable the creation of a new class of flash lidar where each pixel
is served by its own dedicated laser-detector pair. This could provide the advantages of a flash lidar—no moving parts and the ability to capture an entire frame
all at once—without the range sacrifices of conventional flash lidar.
Fundamentally, Ouster's strategy is to piggyback on the industrial base of the consumer electronics sector, which was already using VCSELs for computer mice, the range-finding features of smartphone cameras, and other uses. Pacala argues that VCSELs still have significant room for improvement in terms of brightness, cost, and energy efficiency. As the broader industry invests to improve VCSEL (and SPAD) technology, Ouster will automatically reap the benefits.
Blackmore
Beam steering: Mechanical scanning
Distance measurement: Frequency-modulated continuous-wave
Wavelength: 1550nm
Like Ouster, Blackmore hopes to piggyback on the sophisticated infrastructure of the larger semiconductor industry. But its focus is on the industrial base of the optical communications industry rather than consumer electronics.
While they're superficially different products, lidar and optical networking gear actually have more in common than you might expect. They both send out information encoded in light, then capture the light later and extract information from it.
Blackmore's optical layer is built on standard optical fiber communications components,
the company says on its website. By leveraging decades of development in optical fiber communication, we can say with confidence that our designs are scalable and reliable.
In most other respects, Blackmore's lidar looks remarkably different from Ouster's (and Velodyne's). Rather than spinning around 360 degrees, Blackmore's lidar is fixed in place, with a field of view 120 degrees across and 30 degrees high. Blackmore's lidar uses the frequency-modulated continuous-wave (FMCW) approach to measure distances, which allows the units to measure velocities as well.
Blackmore unveiled a powerful new lidar product a few weeks ago at CES. It initially costs around $20,000 for a single unit and boasts impressive specifications. The company hopes to steadily bring those costs down over time.
Baraja
Beam steering: Spectrum scan lidar
Distance measurement: Amplitude-modulated continuous-wave
Wavelength: 1550nm
Baraja is one of the most intriguing lidar startups I talked to over the last year—and also one of the most secretive.
Most stationary lidar sensors have a field of view of 120 degrees or less, which means that you need to buy four or more lidar units in order to provide 360-degree coverage. That can get expensive, and it also requires putting a bunch of fragile electronics near the edges of the vehicle where they can easily get damaged.
Baraja's big idea is to move all of that fragile gear into the car's trunk. This central processing unit is connected by fiber optic cables to four cheap, durable sensor heads that can be placed on the outside of the vehicle.
In an interview with Ars last summer, CEO Federico Collarte told me that the four sensor heads contain basically silica glass. They're cheap, they're reliable, they withstand the elements very well. If you have a fender-bender, just replace the sensor head.
It's an appealing vision. The only problem is that I can't figure out how it could work—and I couldn't convince Collarte to explain it to me in much detail.
Baraja describes its lidar as spectrum scan lidar,
which means that it steers its lasers by varying the frequency of the laser and then passing it through a prism. That makes perfect sense as a way to steer a beam in a single dimension—but it's hard to see how you could achieve two-dimensional beam steering with this method.
When I asked Collarte about this, he had this to say: For the second dimension, we're using the same concept of spectrum scanning. We do have a mechanical aid in the second dimension.
He adding that the mechanical aid
was not mirrors
and not spinning the laser.
He said that it uses the same prism-like optics—something we're still keeping secret.
Baraja is also the only lidar company we've talked to that uses an amplitude-modulated continuous wave approach to measure distances. Collarte told us that one advantage of the AMCW approach is that it doesn't require the high energy levels of a single pulse.
Some optical components can be damaged by big power spikes, so avoiding them gives engineers the flexibility to use a wider range of design options—potentially allowing for cheaper, more reliable technology.
Collarte says that (like Blackmore), Baraja is aiming to translate components and technologies from optical telecom,
where massive economies of scale help to keep costs down. Baraja seems to still be fairly early in its commercialization process, but Collarte says that, in volumes of hundreds of thousands of units, the company expects to be able to get costs down to the low hundreds.
Quanergy
Beam steering: Optical phased array
Distance measurement: Time of flight
Wavelength: 905 nm
Quanergy got a lot of hype three years ago when it announced a solid-state lidar product that would cost less than $250 at production volumes. But critics say Quanergy has not been able to deliver on these promises.
Quanergy seems to be really struggling to get range out of their sensors,
said Sam Abuelsamid, an industry analyst at Navigant, in a recent interview.
Quanergy is one of the few companies to make lidars with optical phased array technology. Ars explained this concept back in 2017:
A phased array is a row of transmitters that can change the direction of an electromagnetic beam by adjusting the relative phase of the signal from one transmitter to the next.
If the transmitters all emit electromagnetic waves in sync, the beam will be sent out straight ahead—that is, perpendicular to the array. To direct the beam to the left, the transmitters skew the phase of the signal sent out by each antenna, so the signal from transmitters on the left are behind those of transmitters on the right. To direct a beam to the right, the array does the opposite, shifting the phase of the left-most elements ahead of those farther to the right.
This technique has been used for decades in radar systems, where the transmitters are radar antennas. Optical phased arrays apply the same principle for laser light, packing an array of laser emitters into a space small enough to fit on a single chip.
If Quanergy could get this technology to work well, it could have a number of advantages. With no moving parts, the purely solid-state design could be cheap, rugged, and versatile. Much like AEye, Quanergy's lidar can execute flexible, software-defined scanning patterns and dynamically trade off between resolution and refresh rate.
But Quanergy doesn't seem to have gotten much traction in the market. In a November interview, CEO Louay Eldada said that we are hitting our milestones—we are on track.
But there reasons to doubt this. For example, Angus Pacala was a Quanergy co-founder before he left to start Ouster in 2015.
Abuelsamid points to Quanergy's recent focus on using lidar for industrial security—an application that doesn't require as much range self-driving cars. Eldada told me that Quanergy now has a more conventional mechanically-steered lidar product targeted at the security market.
Cepton
Beam steering: Proprietary micromotion technology
Distance measurement: Time of flight
Wavelength: 905 nm
Fully self-driving cars are one of the most demanding applications for lidar, and so far I've largely focused on products aimed at that market. But Cepton is an example of a well-respected lidar maker that's primarily aiming to get its technology used for advanced driver-assistance systems (ADAS). Today's ADAS systems use radar and cameras for basic lane-keeping and dynamic cruise control functions. But automakers are widely expected to incorporate lidar into future vehicles to help enable more sophisticated ADAS systems.
The problem is that (as we've seen) the best lidar systems still cost tens of thousands of dollars—and could still cost thousands of dollars even at production volumes. So companies like Cepton are aiming to build mid-range lidar systems that are affordable enough to incorporate into cars in the next few years.
Indeed, when I asked Cepton CEO Jun Pei last July about the kind of long-range lidar that's needed for fully self-driving cars, he was dismissive of the market, saying he didn't expect customers demanding the units in large quantities any time soon.
Instead, Cepton has focused on the ADAS market, where high-volume lidar deals have already started to be signed. Cepton argues that it has a cost advantage over its rivals.
We're the only company that's capable of selling our lidar below $1,000,
Pei said. And last summer, Cepton announced a deal with Koito, a Japanese company that's one of the world's biggest suppliers of automotive headlights, to incorporate its lidar technology into its headlight designs. This means that if an automaker decides that Cepton's lidar meets its other requirements, it will be able to seamlessly add lidar capabilities to its vehicles.
Pei told me that Cepton's beam-steering micromotion technology is unique to the industry. Traditional MEMS systems use a tiny, mechanically-steered mirror to redirect light. But Pei says Cepton uses a very proprietary optical design that eliminates the mirror is but still capable of imaging with high definition.
He also described it as a small vibratory system that works like a loudspeaker
—but he declined to provide full details about how it worked.
Innoviz
Beam steering: Mechanical scanning
Distance measurement: Time of flight
Wavelength: 905 nm
Like Cepton, Innoviz has focused primarily on scoring large-volume deals with automakers. They've been hawking affordable, medium-range lidar units suitable for use in ADAS systems. And they've been extremely successful at it.
Last April, BMW announced plans to incorporate Innoviz lidar into its cars in the 2021 model year. Also involved in this partnership is Magna, a well-known supplier that will help to provide the logistical muscle required to get a production-ready part into thousands of vehicles.
Automakers are experimenting with lots of different lidar technologies, and so lots of lidar companies can boast of having some kind of deal with car OEMs. But the BMW deal seems to set Innoviz apart from most of its rivals—since BMW seems to be serious about incorporating Innoviz's gear into shipping cars, not just buying units for evaluation or in prototype vehicles.
Car production involves long lead times, so a deal like this will keep a startup like Innoviz busy for several years, and of course scoring one deal could help Innoviz score additional deals in the future.
Innoviz is widely respected in the lidar industry for solid execution.
Innoviz definitely seems to be in a good place,
Abuelsamid told me. I just interviewed Don Walker, the CEO of Magna, the other day. He seems to be very bullish on that relationship.
The BMW deal will presumably be used for ADAS applications, but Innoviz has ambitions in the fully autonomous sector of the market too. Its latest lidar, called InnovizOne, boasts a solid 200 meter range at 50 percent reflectivity and a 120 degree field of view.