Lidar: the Holy Grail of self-driving sensors
Technology companies race to drive down prices for autonomous cars to cruise
Giant tech companies are fighting over the technology in court. Startups around the world are racing to develop new versions of it. And engineers say it is essential to making autonomous cars safe. The obscure object of desire: lidar. “We believe it will be the basis for autonomous driving,” said Guillaume Devauchelle, who oversees innovation at Valeo, a major parts supplier to automakers.
The technology, which uses nearinfrared light to detect the shape of objects around it, is the centrepiece of an intense court fight in California between Uber and Waymo, the selfdriving business operated by Google’s parent, Alphabet. In the case, Waymo accuses a former Google engineer, Anthony Levandowski, of stealing trade secrets about the company’s lidar designs for his own startup — which he sold to Uber for nearly $700 million (U.S.).
Lidar — pronounced LIE-dar — is shorthand for light detection and ranging. It is a type of sensor that is at the heart of many autonomous car designs and is critical to several worldwide high-resolution mapping efforts. The same technology is used to delineate terrain from airplanes and detect speeding violations.
The advantage of lidar is that it can generate precise three-dimensional images of everything from cars to trees to cyclists in a variety of environments and under a variety of lighting conditions. While autonomous car designs use numerous sensors, including ultrasonic, radar and video camera components, lidar has unique abilities. Unlike cameras, lidar cannot be fooled by shadows or blinded by bright sunlight.
The biggest hurdle to widespread lidar adoption is an economic one, and that is where the battle is being waged.
When Google initially started its autonomous vehicle research eight years ago, the lidar sensors it used cost roughly $75,000. Those sensors were made by Velodyne Lidar, an industry leader. Velodyne declined to say what the current pricing is for such systems, but Waymo’s chief executive, John Krafcik, said in a recent presentation that his company had reduced the cost of its lidar system by 90 per cent.
But even at $7,500, such systems are seen as too expensive to meet automakers’ demands.
“Car companies want it to cost $100 and perform 10 times better, be smaller — and very reliable,” said Omer Keilaf, chief executive of Innoviz Technologies, a lidar developer based in Israel. “So there’s a big vacuum in the industry right now.”
The race to fill that void is largely focused on producing solid-state lidar systems, which would shrink the size of the sensors, eliminate moving parts involved in the optical mechanisms and enable the kind of mass manufacturing that could bring costs down, said Hongbo Zhang, a research associate at Virginia Tech who is working on a lidar design. Established automotive suppliers, such as Velodyne and Valeo; technology companies like Waymo and Uber and relative newcomers like Innoviz, LeddarTech and Quanergy all have their sights set on making less expensive sensors.
Luminar Technologies, a lidar company that recently came out of stealth mode, is focusing on improving the performance of sensors by extending the effective range of lidar past 200 metres. (Top-of-the-line sensors now have a range of 120 metres.) Austin Russell, Luminar’s chief executive, said the company accomplished the longer range by using a more sensitive receiver, as well as a more powerful light output that remains safe enough to avoid damaging people’s vision.
Velodyne, which says it is the only current third-party lidar supplier for fully autonomous vehicles now being tested, is well aware that startups are gunning for its business. Velodyne is working on its own solidstate Velarray lidar sensor, said Mike Jellen, Velodyne president, and it plans to start mass-producing them next year in a 200,000-square-foot factory in San Jose, Calif.
While there has been considerable speculation about how Velodyne will face the potential price competition, Jellen declined to estimate how much the new sensors might cost, saying only that a complete lidar sensor package for future vehicles might be priced in the “low thousands” per vehicle.
“The problem is that we’re still in AI learning mode and only buying in quantities of a couple thousand,” said Jellen, referring to artificial intelligence. “The problem is, the market isn’t there yet.”
There are also some technical kinks that need to be ironed out with new systems. Different versions of lidar paint images of the world around them in different ways, said Zhang at Virginia Tech. That means different lidar sensors from different companies cannot simply be exchanged for one another on an autonomous BMW or Ford vehicle. How a car is trained and learns will partly depend on the specific type of lidar used.
Despite these challenges, Keilaf of Innoviz and many others say there will not be fully autonomous cars without lidar — and it will have to be cheap. It will also have to meet exceptionally high standards.
“An autonomous vehicle that’s 99-per-cent safe won’t be good enough,” said Russell of Luminar. “This is mission critical. You can’t afford to miss a single object because that object could be a person.”