Scoping Out the Road to Automotive Lidar

Much of the buzz in lidar technology revolves around the automotive market, and how onboard lidar will empower an autonomous-vehicle future. But what will today’s lidar developers need to do to succeed in the automotive game?

In the first installment of the Visionary Speakers series at the 2018 Frontiers in Optics/Laser Science meeting in Washington, D.C., Jan-Erik Källhammer, the director of visual enhancement and cognitive systems at the US$2 billion automotive-electronics supplier Veoneer Inc., offered a reality check on what it takes to survive and thrive as a supplier to the auto business. He drew in particular on his experience in a very similar area, night vision technology.

Common traits

While the tone of his talk was cautionary, Källhammer himself certainly isn’t bearish on supplying the automotive market. Recalling the first product that he worked on, an inflatable curtain airbag to protect occupants in side-crash collisions, he noted that the product was initially greeted with skepticism regarding its potential usefulness, with demand pegged at no more than 100,000 units. By last year, production had reached 50 million units. “If you’re good at automotive,” he said, “it’s very profitable.”

So how does one get good at automotive? Källhammer spun a tale that’s instructive for lidar developers based on another product he helped develop for the auto industry, night vision technology. “Night vision is interesting for lidar, because it shares common traits,” he said.

Among those traits, he maintained, are a too-high initial cost; a requirement for a raft of new technologies that the auto community isn’t accustomed to; the need to bring numerous new components to industrial scale quickly; and unclear timing. Perhaps most intimidating, Källhammer noted, is that the first year of automotive production for lidar technology will likely need to be higher than the entire accumulated historical production up to that point. “It’s a big leap to make that happen,” he said. “It doesn’t do it by itself.”

Lessons from night vision

The big driver for automotive night vision technology is avoiding pedestrian fatalities at night—but the case for the technology also encompasses, Källhammer pointed out, avoiding accidents with animals, which can be deadly to the human driver. That raised a variety of development issues for night vision that may ring true to today’s autonomous-vehicle crowd—in particular, the need for the system to differentiate between different sizes and kinds of targets, and even to differentiate living from inanimate objects. (“The algorithm,” Källhammer quipped, “is ‘if rock moves, then not rock’.”)

But the real complexities of developing night vision, he continued, were those that emerged as the system was being tested on specific vehicles. One surprise, for example, was that the protective window over the sensor, made with germanium, proved highly susceptible to abrasion and stone impacts—which spelled trouble in some markets, and which ultimately required that the camera be moved to a different part of the vehicle.

Heat management was another unexpected headache; it turned out that backflow of hot air from the engine at traffic stops was boosting the temperature inside the camera at a rate of 6 °C per minute. “That’s bad news for an infrared detector,” he said—and, eventually, it forced a complete sensor redesign.

Still other obstacles included the need for the part to be installable in a minute or less, to fit within auto-assembly-line timescales; and a byzantine array of export controls and regulations, especially in the U.S., that constrained the ability to get parts from certain suppliers. “Export license approval [from the U.S. government] to Germany and points beyond took two years,” said Källhammer. “You have to plan for that.”

Mind the core—and the non-core

For today’s lidar developers, Källhammer suggested, one lesson is that it will be important to focus early and carefully on the core requirements of automotive OEMs—but also on some non-core functions, ranging from mounting location to ambient temperature and vibration to export rules. That’s a lot to have on one’s plate—especially since doing lidar at all in a fast-moving vehicle is no easy trick. “Lidar has to handle all kinds of disturbances,” he noted—rain, fog, dust, smog, snow and the confounding effects of sunlight. “The question is how much degradation can you allow. That determines how much power you need to throw at the problem.”

One tough issue, Källhammer observed, will be achieving the right balance between the range of the lidar system and its resolution. A recent white paper from the company First Sensor, he said, suggests that “fast-driving cars” in an autonomous-vehicle context will need to be able to see at least 150 meters forward, and to detect objects down to 10 centimeters—a vertical resolution of some 0.038 degrees. And, he said, “people then also say that you need to do it for a hundred bucks. And suddenly things become very difficult.”

To license, sell or build?

Finally, beyond technical and cost hurdles, lidar makers must also be woke to the challenge of supplying components to the automotive industry—a demanding and savvy community of buyers, according to Källhammer—and to the significant hurdles in scaling up manufacturing while keeping costs in line and maintaining demanding design tolerances and specs.

In working to overcome these and other challenges, Källhammer suggests, the key will likely be for the lidar developer to focus on its core strengths in a highly competitive market. For some, he said, “it may be more advantageous to license your IP to a more skilled manufacturer—or just to sell your core technology.” Others, he adds, may stay in the manufacturing and production game, but may find it beneficial to cooperate with a tier 1 supplier, to help the transition to volume production.

Above all, based in particular on his experience creating night vision technology for the automotive industry, Källhammer stresses the need for a proactive approach to uncovering problems. “You have to visualize the problems, and find them,” he said. He compared the process of finding problems with a landscape after a flood: “When the water level is high,” he said “the problems are hidden and you have to seek them out. As the water level is reduced, new problems are revealed.” And, he added, be prepared to continue the process of finding and solving problems indefinitely. “You never get to mission accomplished.”