Technology will complement cameras and radar in autonomous vehicles.
Fully autonomous vehicles of the future will depend on a combination of different sensing technologies – advanced vision systems, radar, and light imaging, detection, and ranging (LiDAR). Of the three, LiDAR is now the costliest part of that equation, and there are worldwide efforts to bring down those costs.
Mechanical LiDAR units are currently available, priced in the hundreds of dollars. Those figures have to come down to make the volume adoption of the technology possible in the cost-conscious automotive industry.
Along with the cost factor, LiDAR vendors must be able to show the high performance and reliability of their products. It’s not good enough to have 99% reliability for advanced driver-assistance systems and automated driving. In the safety-critical aspects of automotive manufacturing, the equipment has to demonstrate the “six nines” – 99.9999% reliability.
The importance of advanced technology in automotive vehicles cannot be overstated. Intel’s proposed $15.3 billion acquisition of Mobileye, a vision systems vendor based in Israel, is a case in point. The chipmaker and Mobileye teamed up last year with BMW to collaborate on autonomous-vehicle technology.
LiDAR is a key component of that technology, and investors are opening their wallets for startups working on this technology. Blue-chip investors last month put $10 million into TetraVue, a LiDAR startup in Carlsbad, Calif. Investors include Foxconn, Nautilus Venture Partners, Robert Bosch Venture Capital, and Samsung Catalyst Fund.
Autonomic, a self-driving software startup located in Palo Alto, Calif., has raised around $11 million from Ford Motor and Social Capital. The four co-founders previously worked at Pivotal Labs.
Technavio forecasts the worldwide automotive LiDAR sensors market will see a compound annual growth rate of more than 34% up to 2020. The market research firm estimates the automotive LiDAR market was worth $61.61 million in 2015, with most of the spending in the Europe/Middle East/Africa region and in the Americas.
The company has a report available, Global Automotive LiDAR Sensors Market 2016-2020, published last June, and it will be updating that report during the third quarter of this year.
“LiDAR technology in automotive industry is witnessing rapid evolution, both in terms of technical advancement and market dynamics,” says Siddharth Jaiswal, one of Technavio’s lead industry analysts for automotive electronics research.
Among the key developments cited by Technavio:
1. Cost reduction in an effort toward economies of scale. LiDAR manufacturers are working on reducing the cost of the system by employing efficient processing techniques, and in certain cases positioning products as per customer segments. “For instance, the price of the Velodyne LiDAR unit that is used on Google’s self-driving car is a 64-beam Velodyne HDL- 64E priced at $80,000,” Jaiswal said. “Velodyne also offers 32-beam and 16-beam LiDAR units at $40,000 and $8,000 respectively, which can be used for economical projects. We expect LiDAR technology to follow a similar path of ‘radar’ in the automotive industry, where cost played a crucial role in market adoption. Hence cost is a key focus area for the players.”
2. Compact design. Velodyne’s first LiDAR sensor, released in 2005 was so big and heavy—it weighed about 5 kilograms—that it had to be placed on the roof of the car. The weight is now less than a kilogram, and a solid-state version is compact enough to fit inside the car.
3. Sensor fusion. The technological trend of combining imaging sensors with LiDAR has been a popular technological research topic for over a decade. The data output becomes more reliable if the fusion results in confirming the output of one sensor by validating against the other sensor type. But if the validation doesn’t prove the results of one sensor against another, it makes the system unreliable.
4. Use of LiDAR beyond automobiles in road asset management. Traffic Speed Road Assessment Condition Surveys (TRACS) were introduced on the trunk road network in England in 2000. The U.K. Highways Agency conducts routine automated surveys of trunk road pavement surface condition under the TRACS survey. LiDAR is used to measure distances from the sensor head, and potentially can deliver measurements of objects much further from the survey vehicle than TRACS surveys.
“LiDAR is at a very lucrative position among the autonomous driving sensor suites,” said Jaiswal. “A 360-degree map is its key differentiator from other sensor technologies, and its capabilities with respect to detection of objects and even during the complete absence of light has set its place among OEMs. Also, the evident fall in price of the most expensive device of the autonomous vehicle, the LiDAR sensor unit, is likely to drive the adoption of automotive LiDAR sensors. For instance, Velodyne introduced in 2016 its new LiDAR sensor, the ULTRA Puck VLP-32A. It is claimed to be the most affordable LiDAR sensor capable of addressing vehicle automation levels 1-5 as defined by SAE, and is also very compact compared to the industry’s previous product versions. Because of the solid-state architecture, the sensor is small enough to be mounted on to exterior mirrors while extending the 3D sensing range to 200 meters (656 feet). Velodyne has set target pricing of less than $300 per unit in automotive mass production quantities—a significant cost reduction from the $7,900 per unit of Velodyne’s previous compact LiDAR.”
Moreover, LiDAR can be developed using mature semiconductor process technologies. technologies, and the solid-state version has no moving parts.
“LiDAR is perceived as a key technology for accurate 3D mapping, vehicle awareness, navigation,” said Pierre Cambou, imaging activity leader at Yole Développement. “First there is a race for performance and durability, through the use of short-wave infrared (SWIR) diodes, avalanche photodiode or single-photon avalanche diode. There is also a huge effort in cost reduction. This is mainly trying to make the LiDAR solid-state, through steerable lasers, MEMS micromirrors, or detector arrays.
But Cambou noted there are different approaches to autonomous driving, and LiDAR isn’t essential to all of them. “LiDAR is a fundamental piece of equipment for autonomous vehicles, which I would rather call robotic vehicles. There will be many levels of autonomy. LiDAR might be necessary for city autonomous emergency braking, probably in conjunction with radars and cameras. This multimodality approach is well-defined now. Nobody really questions it anymore.”
And LiDAR’s market will increase as prices drop, from about $600 million today to $1.2 billion over the next five years. “Today there are three entry points in automotive: $3,000, $300, and $30,” he said. “Cameras are currently at the $30 price point and LiDAR is at $3,000. The goal for the LiDAR players is to lower the cost and reach the $300 target without sacrificing too much of the performance. We will see such LiDARs entering the market, probably using solid-state approaches, in the next three years.”
That is a small fraction of the overall vision sensor market. “The consensus is there is almost the same revenue for automotive radar and automotive vision today, but vision is 50% forward ADAS and 50% park assist,” Cambou said. “We have reached $1 billion of automotive vision sensor value in 2016 and the growth is 24% CAGR. The horizon is $7.3 billion in automotive vision sensor revenue by 2021.”
What works, what doesn’t
Amin Kashi, director of ADAS and Automated Driving at Mentor Graphics, a Siemens business, said that interest in LiDAR began more than a decade ago due to the high cost of radar sensors at the time, which cost about $500 apiece. LiDAR sensors were extremely expensive then, at up to $260,000 per unit.
“Three years ago, you saw a number of companies or startups beginning to invest in and look into the LiDAR space,” Kashi said. “Every major Tier 1 somehow has started investing or acquiring companies in the LiDAR space.”
That includes companies such as Continental and TRW. Kashi previously worked at Quanergy Systems, which developed a mechanical LiDAR sensor and is working on a phased-array LiDAR sensor. Quanergy’s solid-state LiDAR sensor goes for about $250.
Meanwhile, Mentor Graphics, a Siemens company, is providing hardware, software, and design services to OEMs and Tier 1s addressing LiDAR. “We’re also providing software IP that their sensors can run. At the end of the day, all of the sensors have to somehow be fused. There needs to be a processing platform or system that takes all of this different information and makes it available for the decision engine. That’s where our interest is.”
Cameras, LiDAR, and radar are complementary to each other, providing redundancy for the deficiencies of each technology, he said. That’s critical because LiDAR can be less effective in fog and low clouds, dust storms, heavy rain, and heavy snow.
“You still have to have very good resolution for the sensors you use for your autonomous vehicles,” he noted. “There are a lot of companies working on LiDAR technology, a lot of startups, and they have very compelling concepts. The interesting thing is going to be is to see if the road to commercialization is going to be successful. Some of these are very imitative, but it’s a big challenge going from a great concept into an automotive-grade sensor. And there is a lot of investment associated with that.”
Making comparisons between the various LiDAR technologies isn’t always straightforward, though, and it’s not made any easier as competition heats up.
“There’s lots of misleading information out there,” said Louay Eldada, CEO of startup Quanergy. “You have people who do traditional mechanical LiDAR—big, spinning mechanical LiDAR that’s used in helicopters—and they call themselves hybrid solid-state because the semiconductor content is non-zero. That’s just deception.”
Such products have one small chip in a bucket-sized product, according to Eldada. “In the automotive space, no one is still using mechanical LiDAR. We believe strongly that our solid-state LiDAR is by far the most exciting development in this space.”
Quanergy last year received $90 million in Series B funding, bringing the total of its private funding to about $150 million and valuing the company at more than $1 billion. Delphi Automotive, GP Capital, Motus Ventures, Samsung Ventures, and Sensata Technologies invested in the Series B round.
XenomatiX, another startup, also focuses on solid-state LiDAR. “Startups are taking the lead in development that is considered to be essential for automated driving,” said Filip Geuens, CEO of the Belgium-based company. “There are huge investments and expensive acquisitions by some of the big guys to get the sensors and software required for automated driving. Most of these companies, technology-wise, are going in the same direction. We expect they will all hit essential hurdles. We are walking in a different direction and doing things slightly differently, because we believe this is the best way to overcome these hurdles.”
XenomatiX is trying to clear up sensing confusion among LiDAR systems, with many systems utilizing direct time-of-flight sending out one beam of light or one flash of light, Geuens said. “The direction we are taking is to send out thousands of beams at the same time. It’s quite a challenge. We are also heeding the eye-safety restrictions. That is the most important hurdle that’s the same for all of us. We’re sending out many beams at the same time, and that makes it even harder. The upside is it makes the system so much more reliable in real circumstances where multiple LiDAR systems are operating at the same time.”
Some companies assert that cameras and radar are sufficient for automated driving. Geuens doesn’t believe that. He said that driving a car involves a 3D world, and LiDAR is essential for sensing in all directions.
One big issue in the industry is the push-and-pull between the OEMs and the Tier 1s. OEMs traditionally expect Tier 1s to bring them the advanced technology they need, while Tier 1 companies need proven technology before presenting it to the OEMs. According to numerous industry insiders, the vendors of automotive components don’t want to spend massively on R&D without OEM commitments to volume purchase orders.
Intel’s pending purchase of Mobileye is “a big step forward” in bringing high-technology products to the automotive industry, Geuens said.
But the race toward autonomous vehicles, and the amount of technological innovation required to get there, is bending some of the previous approaches. “Right now, LiDAR technology as a whole is kind of morphing,” said Jean-Yves Deschênes, president of Quebec-based Phantom Intelligence. “That morphing is caused by the automotive industry.”
Five to 10 years ago, LiDAR was primarily used for architectural, mapping, and military purposes. The units were typically huge, unwieldly devices with many mirrors.
“A lot of people are looking for a solution,” he said. “Recent research and companies we hear a lot about right now are trying to replace those mirrors. We produce the scan LiDAR principle, by using MEMS mirrors, beam steering, whatever. A lot of mapping is going in that direction. We believe strongly at Phantom Intelligence that the solution lies more in flash LiDAR technology. Flash LiDAR is pretty much more of an analog to a 3D camera. Instead of having a narrow beam being geared to progressively sweep that field of view to recreate the image, you flash the image with laser pulse over a large surface and use multiple pixels to reconstruct the image.”
LiDAR’s disadvantage are the echoes coming back to the sensor, noted Deschênes, who favors what he calls more intelligent signal processing. He predicts there will be five levels of autonomous driving, with fully autonomous vehicles rolling out in 2025 and widespread adoption of the technology in 2030.
Put in perspective, LiDAR is an well-known technology that has finally found a lucrative market application.
“The principle of LiDAR – the light sent through the pulse and echo of time-of-flight – has not really changed,” said one industry source. “The physics have not changed ever since its invention, for the past 40 years or so. The evolving changes are more in the components and system integration. There’s no fundamental principle change.”
Flash LiDAR has been in development for the past five years, the source noted, likening it to a CMOS image sensor. “This is an area to watch for—the flash LiDAR technology. It promises a very low cost of solution, not necessarily high performance.”
Kevin Watson, senior director of product engineering at Redmond, Wash.-based MicroVision, a publicly held company, disagrees. “I don’t think that’s going to go anywhere,” he said of flash LiDAR. “For many years, the Holy Grail of LiDAR sensors we thought to be a MEMS mirror-based laser scanner, because they’re super-small, relatively inexpensive to manufacture in great quantity, and very reliable. They’re small enough to hide several around an automobile.”
Watson calls LiDAR “the most important sensor” in automotive electronics. “Vision systems are great, but they’re a totally passive system. LiDAR is active.”
But LiDAR also has its limitations. Radar can recognize a wall and has a longer range and it also works in fog, while LiDAR and vision can be confounded. Achieving Level 4 autonomy, the next-to-highest level, is “a ways off,” said Watson, adding that may not be realized for a decade. “It’s a very, very tough problem. It’s just a lot of work.”
Teaching Computers To See
First of two parts: Vision processing is a specialized design task, requiring dedicated circuitry to achieve low power and latency.
What’s New In Connected Autos
Internet of Things technology will be crucial to automobiles, but connectivity comes at a price.
What Can Go Wrong In Automotive
Experts at the Table, part 2: Understanding security risks, ECUs vs. SoCs; dealing with an explosion in data.
Sensors Enable ADAS
Advanced driver assistance requires a careful and complex balance of hardware, software and security.
Seeing The Future Of Vision
Self-driving cars and other uses call for more sophisticated vision systems.
Prioritizing Vehicle Data Traffic
Challenges grow for classifying and tagging huge volumes of data from connected cars.