Velodyne Pads Lead in Lidar Derby

　　Velodyne LiDAR is still lapping the field in an embryonic but increasingly competitive and complex lidar market, an advantage that grew this week when it unveiled a new 128-channel lidar sensor. The VLS-128 lidar unit boasts range (distance) and resolution that other lidars currently on sale have not been able to offer.

　　Velodyne’s VLS-128 is ideal for “high-speed highway driving, as it senses objects in farther distance,” said Anand Gopalan, CTO of Velodyne LiDAR. Moreover, it can capture “a rich set of data, high-resolution enough for object classifications without using a camera,” he added.

　　To pass as an effective sensor technology for highly automated vehicles traveling at 70 miles per hour on a highway, a lidar needs to be able to see at least 200 to 250 meters ahead, recognize that there’s an object out there, and determine what it is.

　　Compared to its own previous model (HDL-64), Gopalan said Velodyne’s VLS-128 can see objects three times farther (300 meters) and in three times resolution (0.1 degree).

　　Armed with these impressive specs, Velodyne hopes to set itself apart from traditional lidar sensors that are too low-quality or used primarily on a development platform. However, Velodyne’s own lidars, thus far, have been primarily used to create highly accurate 3D maps of their surroundings by bouncing laser beams off of nearby objects.

　　Velodyne’s VLS-128 is coming to the market as the competition for lidar is heating up, with the development community still wrestling with a host of newly emerging technologies.

　　Big car OEMs are snatching up lidar technology companies. For example, Ford bought Princeton Lightwave just last month, General Motors acquired lidar company Strobe Inc. also last month, and Continental got the lidar business from Advanced Scientific Concepts (ASC) last year, explained Akhilesh Kona, senior analyst, Automotive Electronics & Semiconductors at IHS Markit.

　　On one hand, the industry sees on the horizon a new laser emitter technology — above 1,400-nm wavelength. The laser in the new wavelength promises to bring to lidars higher resolution and longer range, said Kona. Princeton Lightwave, Continental (through its acquisition of ASC), and Luminar Technologies are all working on new laser emitter technology, he added.

　　On the other hand, technology suppliers continue to improve the durability, size, and cost of their lidars by developing a variety of beam-steering technologies, said Kona. They range from mechanical to MEMS and solid-state.

　　Lidar vs. other sensors

　　It’s important to note that the consensus among automakers is unequivocally the need for multi-modal sensors in autonomous vehicles. Velodyne isn’t claiming that lidars will replace other sensors already in automated vehicles. Rather, the company is pitching its improved lidars “for the safety and redundancy of the autonomous vehicles’ compute function.”

　　In assessing the performance improvements in lidar technologies, however, it’s helpful to understand how a lidar stacks up against other sensors such as vision and radar.

　　Phil Magney, founder and principal advisor for Vision Systems Intelligence (VSI Labs), noted, “Lidar’s advantage over other sensors is for every point you have a precise distance measurement. However, the problem with lidar is its relatively low resolution or its ability to distinguish colors.”

　　Velodyne’s VLS-128 lidar compensates the traditional weakness of lidars by increasing the resolution, so the new VLS-128 can classify objects as the company claims, explained Magney.

　　“Cameras, on the other hand, have high resolution, so their ability to classify an object is much better than lidar. But cameras don’t have the precision on distances,” noted Magney. “To deal with this, you usually fuse radar measurements to the objects from the camera and you end up with acceptable precision for some applications.”

　　What about radar? “Radar offers precise distance measurements but almost no resolution,” he said. “The radar will know there is an object there and its exact movement and velocity with respect to the vehicle. Newer radars (millimeter-range radar) do have resolution and can pick up on multiple points of an object and even classify it.”

　　Evolutionary path

　　There is no question that Velodyne has made significant improvements on its own mechanical-based lidars.

　　As Magney sees it, “Velodyne is state-of-the-art when it comes to lidar, but their market is still the development side, where unit cost is not a big issue.”

　　Magney observed, “While 360-degree laser scanning is desirable for development and making high-definition maps, a reduced field of view (FOV) is viable for production vehicles, as is the case with the new Audi A8, which uses the Valeo Scala unit. This is a forward-facing device and is also an intelligent lidar device that produces object data.”

　　While Velodyne is at the high end, Magney believes that a number of low-cost lidar devices will likely make their way into production vehicles. “The Valeo unit is one example, but others from Quanergy and Pioneer purport similar capabilities,” he noted. “These are devices that cost in the range of a few hundred dollars. There are lower-cost flash devices emerging that cost well under $100, but their functionality is limited by low resolutions.”

　　In predicting new technologies on horizon for lidar, IHS Markit’s Kona laid out three evolutionary phases: mechanical scanning (available now), solid-state scanning (a system in production around 2020), and pure solid-state (after 2020).

　　As shown in the table above, pure solid-state lidars come in different types. They range from basic flash to high-resolution flash, optical phased array and frequency modulated continuous wave, Kona noted. While basic flash is already in production — largely used for ADAS systems — other types of pure solid-state lidar are still in development.

　　No lidar company is married to a specific beam steering technology. Companies like Velodyne and Valeo, for example, are working on both types of lidar — mechanical and pure solid-state.

　　Kona also pointed out that different “use cases” demand different lidars. Primarily for highway driving, very long-range lidars with a narrow field of view is necessary. For city driving, lidars with a broader field of view are critical to see corners at an intersection or detect pedestrians, he noted.

　　Lidar vendors such as Continental and Valeo already offer solid-state flash lidars. But those lidars, whose cost is in the $100 range, have limited resolution and range, explained Kona.

　　In contrast, companies such as Velodyne, Ibeo, and Valeo provide high-resolution mechanical lidars with 360 degrees. Price, however, tends to be high (Velodyne’s puck is $8,000) and size too big, said Kona. Another downside of such high-resolution mechanical lidars is an abundance of moving parts, making them susceptible to vehicle vibration.

　　What OEMS want

　　Velodyne’s claim for VLS-128’s object-classification capabilities has triggered arguments on the architecture of autonomous vehicles. At issue is whether AVs might eventually opt for a central-fusion architecture or a distributed sensor processing model.

　　Velodyne’s CTO told us that many OEMs and Tier Ones are looking for intelligent lidars capable of pre-processing of data that enables object classification. Until now, OEMs had no choice but to fuse raw data from lidar with a camera because no intelligent lidars were available, said Gopalan. The goal of intelligent lidars is to provide a high level of information including object lists, localization, and segmentation, he added.

　　VSI’s Magney noted, “Velodyne now claims that the higher resolution can classify objects well, and so you would not need any supplemental sensors.” He explained, “When configured with the proper classification algorithms, you would have everything you need to get a good-enough environmental model to have a safe deployment. Under this use case, you would not need to fuse the Lidar’s object data with anything else.”

　　Magney, however, disagreed with Velodyne’s claim that theirs is the only lidar that classifies objects. “There are other lidars out there that also couple classification algorithms within the sensor package and are intelligence sensors because of this. Valeo’s Scala unit is an example of that kind of device even though it is a forward-facing device, not 360-degree like the new Velodyne unit.”

　　Asked if the automotive industry is indeed going for a distributed sensor processing route, Magney made it clear, “Not everybody wants processed data.”

　　He said, “Advocates of AI desire raw data rather than processed data. There are hybrids of this, too, where you have partially processed data that is something short of object data.”

　　IHS Markit’s Kona agreed. “There is no one right answer to this.” Noting that some OEMs are already working on their own software algorithms for autonomous driving, he said, “Some prefer raw data from lidars (and other sensors), not the processed data.” But others without their own algorithm development are likely to demand lidars (or other sensors) capable of object classification.

　　Solution for corner cases?

　　In its own press release, Velodyne boasted:

　　With its long range and high-resolution data, the VLS-128 allows autonomous vehicles to function just as well in highway scenarios as low-speed urban environments. It is designed to solve for all corner cases needed for full autonomy in highway scenarios, allowing for expanded functionality and increased testing in new environments.

　　But can lidar solve “all corner cases”? Can an AV equipped with the VLS-128 always recognize that a plastic bag blowing across the road is indeed a plastic bag, not a hard object?

　　Magney noted, “While the VLS-128 will have a very precise 360-degree view of any scene, it is never enough data to solve all corner cases in my opinion.” He added, “Some corner cases will involve occluded objects that the sensors simply cannot see. On the other hand, the precision data that the VLS-128 acquires will be very accurate and be able to pick up on just about everything going on in the scene (short of occluded objects). So from a perception standpoint, the data you get from the new Velodyne unit will be as good as lidar gets.”

　　However, Magney stressed, “You have to keep in mind that coping with corner cases is largely determined by predictive software and not the lidar data itself. However, couple the accuracy of precision lidar with artificial intelligence and you might have your best defense at handling a diverse range of corner cases.”

　　IHS Markit’s Kona agreed. While acknowledging that Velodyne’s claim for “solving all corner cases” might be “marketing spin,” he explained that, given deep-learning algorithms, point cloud information collected by lidar could make it easier to connect the dots.

　　But what about cost?

　　Velodyne refused to disclose how much the new VLS-128 lidar will cost. Velodyne’s lowest cost Puck, VLS-16, costs $8,000.

　　Magney said, “I would say cost is the biggest holdback on lidar right now.” While Velodyne did not say how much the new unit costs, Velodyne’s previous lidar model with 64 channels cost around $75,000, he noted. “So it is probably likely that the new one is going to be too expensive for production cars.”

　　Magney sees the VLS-128 “best applied to development.” He explained, “Most of the time, 360-degree lidars like the Velodyne 64 and 128 are used for making detailed maps, so their use case is limited to development.”

　　However, Magney added that he would not rule out Velodyne as a key supplier to commercial vehicle fleets. Presumably, robotic taxis and autonomous trucks will have a different economic model. “They can support a more expensive sensor package, and they will. As far as the VLS-128 being deployed for these, this may depend on its price,” said Magney.

　　Manufacturability

　　In announcing VLS-128, Velodyne discussed its investment in automated assembly in Velodyne’s San Jose factory.

　　Annual production capacity at the factory will increase to a million lidar units in 2018, according to Velodyne. While the company’s mass production plan is reassuring, it also begs the question as to how complex the production of lidars might be.

　　Kona observed, “In general, manufacturing mechanical scanning lidars is quite challenging. Integrating mechanical mirrors, motors, and other semiconductor components inside a module that should sustain harsh conditions (vibrations, extreme temperatures, etc.) is a formidable task.” But he added that Velodyne has been doing this for over a decade now, as well as working with many OEMs.

　　Kona suspected, “Leveraging on this expertise, Velodyne seems to have a patent in manufacturing lidars (laser alignment and system manufacturing). The in-house expertise at various levels of lidar development and production would give them an edge over other suppliers.”

　　Velodyne’s Gopalan told EE Times that the lidar company has designed its own ASICs to reduce the number of components inside the unit. What used to be a few thousand components (for 32-channel lidar) is down to “a couple of hundred” for the latest 128-channel unit, he said. Those range from analog and mixed-signal chips to A/D converters and digital signal processing. “It’s across the board,” he said.

　　Kona believes that semiconductor technology will ease manufacturing. “Velodyne’s new approach of integrating functions on fewer semiconductor components and using mass-produced semiconductor technologies will accelerate its production volumes,” he noted.