At this spring’s Tesla Autonomy Investor Day event, Elon Musk once again dissed the “fools” relying on LiDAR for self-driving vehicle development — prompting the hashtag “LiDAR” to trend globally on Twitter.
Computer vision and hardware engineers and researchers are familiar with the hackneyed LiDAR debate. Although Light Detection and Ranging has proven its reliability and accurate ranging capabilities, the hardware’s size and cost of some US$10,000 per unit are huge obstacles to its mass commercialization.
The technical question is whether to wait 10 or more years for computer vision research to solve the ranging problem or for cheap LiDARs to hit the market. However, since the business mindset is to focus on short-term feasible technologies, the lack of serious buyers is the real problem for the LiDAR industry.
Defending Doubts from Buyers and Partners
A number of startups launched solid-state LiDARs in 2016-2017, optimistic that unit prices would drop to hundreds of dollars with mass production. Finding bulk buyers to drive demand has however been difficult.
In order to survive, LiDAR companies need to either secure orders from L4 autopilot teams or integrate with auto manufacturers’ supply chains. The L4 research teams can only support a few LiDAR makers, while most traditional auto makers have yet to embrace the tech — a notable exception being the Audi A8’s use of the discreet Valeo SCALA.
Top tier auto parts suppliers such as Bosch, Aptiv, ZFAuto part manufacturers have not exactly embraced the commercialization of L3/L4 self-driving tech such as pricey LiDARs. At the Shanghai Auto Show this April a Bosch rep explained, “Probably by 2020, we will have more than 40 models installing L2 systems, but L3, not in the near future.”
Small autonomous driving startups, who used to partner with LiDAR companies at auto shows, are resetting their agendas. Some are using pre-equipped vehicles or testing other tech products, while many have left the industry altogether. A startup rep told us “car manufacturers are considering many promising sensors for mass production targeting the year 2021… unfortunately, LiDAR is not on the list.”
LiDAR manufacturers have not given up — one developer told Synced that “people who believe in pure computer vision are fools… I’m certain that by 2021 LiDAR can enter mass production because the only barrier now is regulations. After that’s solved, LiDARs will be equipped on green-energy or high-end cars, which will reduce the fraction cost, since retail price of these cars is US$40,000 and above.”
Companies and Their Over-Marketing Problems
As the academic and industry debates continue, one LiDAR poster child is being criticized for over-marketing.
Founded in 2012, LiDAR unicorn Quanergy has completed five rounds of financing totaling more than US$100 million for its solid-state LiDARs. In 2018 however Bloomberg published a feature story, How a Billion-Dollar Autonomous Vehicle Startup Lost Its Way, detailing Quanergy’s technical stasis.
For example, although the startup strove to produce LiDAR products on a strict schedule, some insiders claim the products were not as good as advertised, and that no solid-state equipment meeting Quanergy requirements has yet come off the company’s assembly line.
In addition, the mechanical LiDAR prototype M8, introduced in 2016, was roundly returned by customers. Meanwhile, Quanergy has begun to shift its tech focus away from LiDAR, with plans for example to help develop President Donald Trump’s digital Mexican border wall.
Quanergy is not the only company running into problems. An engineer working on L4 autopilot technology told Synced “testing early products of global LiDAR companies reveals big differences between advertised accuracy, vulnerability, product life, and detection range and the real world performance of the products.”
It was an open secret a couple of years ago that many solid-state LiDAR products at high-profile launch events were actually usable, and only mounted atop cars for show.
Only a Few Companies Will Survive the Winter
The year 2013 saw breakthroughs for LiDAR companies. Quanergy Systems, favoured by the capital market for several years, completed its seed round financing; TriLumina, which provides lighting solutions for the industrial sector, reached out to the automotive field; and close to a hundred companies crammed into the promising market subsection.
By 2019 however market activity had subsided. Now, investors are putting their money into just a few companies. Innoviz, Ouster, and Luminar are among those who have pocketed tens to hundreds of millions in financing rounds.
Moving forward, it is natural that only those LiDAR companies that can garner buyer support will survive. And what will become of the others? “We are not talking about accelerating mergers and acquisitions, but companies dying in this field,” an industry insider tells Synced. “It’s not that the big guys are necessarily better, it’s just that the others are all gone.”
Source: Synced China
Localization: Meghan Han | Editor: Michael Sarazen | Producer: Chain Zhang
Pingback: LiDAR Industry Hits Impasse: Was Elon Musk Right After All? – NewsChest Technology
Setting aside the proposition here–the occupancy of the lidar community–one still must wonder if autonomy will work with any sensor -measured world easily confounded with rain, fog, ice, bug juice, setting sun in your eyes, etc. I don’t know if investors are asking these questions.
I can surely attest that current Tesla’s current iteration of autopilot deals very well with rain, dust, fog, “bug juice” and direct sun, better than I do. I have not had the chance to test snow, however others have, successfully. They are very serious about full self driving, experiencing the progression of the system is the proof.
To add to to that list of questions, how about color. The lidar “point cloud” representation of the world may be great for capturing distance information, but since lidar actively beams light out, it can only use the non-visible spectrum (to avoid turn roadways in a seizure inducing rave party). This greatly hampers object recognition potential. The existing driving infrastructure makes extensive us of color information in the human visible spectrum.
You really don’t have to look much past the poor performance in heavy rain, hail, snow, to discount it’s use as an integral part in a sensor suite.
It’s great for demos that create the illusion of progress to garner investment, not much else. Perhaps an extra layer of redundancy when the weather is nice.
So far 3 for 3 responses and article which are so far detached from reality, it is almost funny. Any vision system such as Tesla’s will not come close to LiDAR in being able to identify noise in inclement weather. LiDAR’s have the ability to get up to quad returns to address wather and dust, vision systems can only rely on algorithms filtering out rain, dust, snow, and doing so adds an enormous amount of noise to the capture and renders edge recognition for data libraries useless.
LiDAR is not there to replace cameras, but rather complement the current suite of sensors.
Cameras are poor at estimating distance, but their high pixel densities make them perfect for classification tasks. Even the current high end lidars can’t match that density for classification. But they sure are more reliable for distance estimation.
The problem, as stated in the article, is that the price point has not been met for auto manufacturers to accept adding that complementary (yet necessary) reliable depth perception.
This is just patently not true.
Of course Elon Musk was right. He gets the timing wrong, but nothing else. Every other time “the experts” have said Musk was wrong, they were wrong. Now all these companies have given Tesla a five year head start on autonomous; gee, I wonder what company will achieve level four and five first?
Meanwhile, stock “analysts” assuming Waymo will win autonomous suggest it is worth $75 billion, or about twice what Tesla’s current valuation is. When Tesla reveals level four or five, the company should immediately be worth three times as much so that seems like an easy way to make some money.
Elan is dead wrong. He’s just stalling, waiting for the cost of LiDAR to come down using a proven CMOS process. In the mean while, he is using his customers as guinea pigs instead of safety drivers, at their peril. It will be years before any vision system has the accuracy as liDAR. Not to mention the hundreds of billions of images that need to be labeled to provide an accurate RCCN library, covering every circumstance, every location, every time of day, every object, every object at different times of day and every season. That is why his engineers have been purchasing LiDAR test units.
People always remember the cameras, but forget the radar, and ultra sonics.
Waymo “winning” autonomous. Spoken like a true outsider who doesn’t work in the industry, and just draws conclusions based on his buddies comments. Waymo recently has had some serious set backs, (public information) according to their engineers. They will be one of the “players” in the robo-taxi sector, but by far not the leader. One of the major OEMs will be. You’ve got to have vehicle plus service to win this business model/market sector. Waymo does not sell any liDAR to other of the 64 California registered test programs. They only provide their liDAR to their own test fleet.
So far 3 for 3 responses and article which are so far detached from reality, it is almost funny. Any vision system such as Tesla’s will not come close to LiDAR in being able to identify noise in inclement weather. LiDAR’s have the ability to get up to quad returns to address wather and dust, vision systems can only rely on algorithms filtering out rain, dust, snow, and doing so adds an enormous amount of noise to the capture and renders edge recognition for data libraries useless.
I found your article speaking from a way, way outsiders point of view and does not reflect in any way what is actually ocuring within the LiDAR marketplace. Sorry to be critical, but your article is not close to getting true, accurate information inside this industry. Your Woodside Capital graphic along with just about all Woodside Capital market information is woefully outdated. This is 2-year old information in a market that is extremely dynamic. Unfortunately, it is going to be impossible for anyone to get accurate information from leading vendors at this juncture because the race is on. this industry is NOT in a slowdown. Orders are actually picking up rather than declining as certain markets such as robotaxi , people movers, short haul delivery and geo-fenced off-road applications are moving to mass market in 2020. Lead times for this ramp up have to be met. Market share leaders by far for LiDAR, based on units sold and revenue include an American company and a Chinese company. The rest are considered at this point “also rans” with less than 10% share. As a resident living in Silicon Valley, all I have to do is drive home from work and count LiDAR placements on every test vehicle to get a good idea who is selling to whom, who is overtaking who and who are the current leaders. I suggest you do the same for your next article.
“Innoviz, Ouster, and Luminar are among those who have pocketed tens to hundreds of millions in financing rounds.” These are by far NOT the only companies. You just haven’t heard about the others.
“Top tier auto parts suppliers such as Bosch, Aptiv, ZFAuto part manufacturers have not exactly embraced the commercialization of L3/L4 self-driving tech such as pricey LiDARs. At the Shanghai Auto Show this April a Bosch rep explained, “Probably by 2020, we will have more than 40 models installing L2 systems, but L3, not in the near future.” Your Bosch and Aptiv connections are not telling you what they are, or are not doing. To say they have not embraced commercialization is not an accurate statement, even with basic public information.
“testing early products of global LiDAR companies reveals big differences between advertised accuracy, vulnerability, product life, and detection range and the real world performance of the products.” This is complete BS! Where did you find this outsider? This may be true with certain companies, but not all. What is inaccurate is your article.
In other words why does startups are experiencing a high success rate and below-average failure rate
Correction: why does…
Lidar is used more of a cross validation. If our leading camera companies don’t have a 100% hit rate on instant focus what makes you think they are on the road?
Lidar is NOT used for validation. The camera portion of the stack is used to validate, adding contextual information to the LiDAR.
New technology tends to ‘reformat’ products (size, appearance, functions). Smartphone good example. Regardless of the vehicle autonomy system or principle, I’d say that it is better to ‘reformat’ cars before any autonomous tech goes in.
1. The less space they require, the more margin there is to evade other road users.
2. The easier the auto-pilot’s task of overseeing the vehicle and its surroundings, particularly if it’s shaped to avoid the dreaded blind spots.
3. If necessary, the human driver can act as a fail-safe; great outside view will boost involvement as well.
LIDAR shows a lot of promise. It’s the gold standard for 3D model accuracy in self-driving cars. But it is definitely in Gartner’s Trough of Disillusionment.
It works as promised, but hasn’t dropped in price as we had hoped. And it is unlikely to scale until prices drops, and that’s a bit of a Catch-22.
Meanwhile, the @pmarca rule that “Software will eat the world” plays in favor or @elonMusk and cameras for autonomous cars.
Camera solutions rely on:
1 already-cheap imaging sensors
2 powerful hardware that benefits from Moore’s Law, & has declining costs over time
3 software (with Marginal Costs near zero)
One thing for sure: Competition is good! Let the different methods battle it out.
Pingback: Semiconductor Engineering - Week in Review – IoT, Security, Autos