Today automotive radar systems are shortsighted. They lack the resolution to distinguish objects close to the vehicle, while discarded soda cans and other metallic objects can cause false alarms that increase the risk of accidents. They also struggle to process information fast enough for highway driving.
At the same time, the radar is critical for the future of autonomous driving. The technology can corroborate what other sensors perceive about the car’s surroundings. They also work in the dark and all weather conditions as well as watch for other vehicles while the cameras and lidar are obscured by rain or thrown off by the sunlight reflecting off a snowy street.
But a new generation of startups is focused on high resolution radar that not only senses the presence of obstacles on the road but also generates rough images of its surroundings, giving it qualities of lidar sensors and without sacrificing bad weather performance. These companies are also targeting longer range radars that can give cars advance notice about potential dangers on the road ahead.
We have been watching a number of automotive radar startups in recent months and summarized what they are doing. Many radar companies are expanding upon systems first used in missile defense and aerospace applications, while others are improving existing categories of radar currently shipped into millions of cars, like frequency modulated continuous wave (FMCW).
The incumbents include Texas Instruments, which last year released a single millimeter wave chip that measures the distance to objects within the width of a human hair and could be used in high resolution systems. NXP – which estimates that its chips shipped in more than 50 percent of all radar modules last year – supplies millimeter wave radar that operates over the 77 GHz band. Traditional automotive suppliers are also toying with high resolution radar.
It's not yet clear that high resolution radar will catch on. Autonomous cars will employ a wide array of sensors – including cameras, radar and lidar – to provide redundancy. The best combination of sensors is still up in the air and who knows what technology will provide the best value. If other sensors improve significantly, high resolution radar could be overkill.
But with the current state of sensors, the technology would likely improve localization. Today, lidar cost thousands of dollars and fumble in fog and snow despite high resolution. The reliability of such sensors has also been called into question because the most accurate ones need to spin mechanically to scan in all directions – at least until smaller and cheaper solid state lidars are developed.
The executives we talked to don't expect high resolution radar to replace lidar, which is considered essential to autonomous driving by almost the entire automotive industry. But many of the startups have built prototypes and could release products within the next two years for an early generation of semi-autonomous cars.
Uhnder. Last month, automotive supplier Magna announced a high resolution radar it had built with Uhnder, a stealth startup based in Austin, Texas. The division of labor between the two companies is not clear and Uhnder declined to divulge more details. The radar is designed for Level 3 to Level 5 autonomous driving and will be available in 2019.
The Icon radar scans the environment continuously to measure the distance, height, depth and speed of objects out to 300 meters. (The term 4D radar comes from these four dimensions). The radar tracks a hundred times more objects than competitive systems and identifies them individually, said Magna. The company also implied that the radar uses MIMO, a technology traditionally used in telecom.
Arbe Robotics. Last year, Arbe Robotics closed a $9 million funding round with plans to improve the simultaneous localization and mapping (SLAM) algorithms in its radar system. The software runs inside an embedded processor that builds a point cloud 50 times per second and localizes the vehicle 25 times per second with a single degree of resolution between objects, the company says.
The software is paired with its synthetic aperture radar, which uses more channels that normal automotive radar to boost resolution. The radar’s range is around 300 meters. The company is targeting Level 4 autonomous cars and the millimeter wave radar will be used in autonomous car pilots within the next two years, according to Kobi Marenko, Arbe’s chief executive.
Ghostwave. Last year, the U.S. Federal Communications Commission opened five gigahertz between 76 and 81 GHz for car applications. The additional spectrum can be used to improve radar’s range accuracy and separation, but millions of cars and the millions of radar inside them could have trouble piling into the same spectrum without causing interference.
Ghostwave’s radar is designed to completely avoid unintentional jamming. The 24 GHz radar can also distinguish between low and high priority obstacles, said chief executive Dean Zody. The radar’s performance is not groundbreaking but the interference capabilities should help it punch above its weight class. It could be built into rear bumpers to replace ultrasonic sensors in collision warning systems.
Metawave. Spun out of the Palo Alto Research Center last year, Metawave has prototyped a radar system that operates over the 77 GHz band. Its main innovation is antenna fabricated out of metamaterials, which exploit tiny repeating structures to electronically steer radar beams to a target. The technology is called analog beamforming, Metawave says, and it works without the bulky and expensive phase shifters used by defense systems to steer beams.
Metawave’s radar system uses deep learning to determine the size and shape of objects around a vehicle. That way, companies could use the system to generate digital road maps using radar instead of lidar or cameras. These radar maps have the potential to be valuable to automakers. Metawave C.E.O. Maya Achour declined to disclose when the startup would release its first product.
Echodyne. Two years ago, the company released its first product to enable drones to safely leave the pilot’s line of sight to survey farms or construction sites. Last year, Echodyne raised $29 million in funding to help it increase production of its compact drone radar system from hundreds of units per year to thousands, improve its radar vision software, and build an autonomous car radar unit.
Like Metawave, Echodyne uses metamaterial antennas to enhance the resolution of its drone radar, which can roughly image and follow large drones at a half mile and aircraft at longer distances. Echodyne’s metamaterials are repeating patterns of copper wire traced over ordinary circuit boards, which can be electronically steered by specialized algorithms.
Oculii. Based in the former auto manufacturing mecca of Dayton, Ohio, the company has developed a 4D radar system that can follow 200 targets simultaneously 200 meters out for mapping and localization. The company optimized its radar system to combine information with video streamed in from cameras around the vehicle. Over the last five years, Oculii has raised $14.3 million from investors including Samsung.
Vayyar Imaging. This company, which recently closed a $45 million funding round, is developing a radar system capable of seeing through walls and mapping everything from individual objects to entire factory floors. Vayyar’s system can also be used inside cars to keep track of whether the driver is falling asleep behind the wheel. The company plans to use the same three-dimensional radar sensor to scan other cars on the road, but we don’t have more details on that effort right now.
Steradian Semiconductor. This company was founded in 2016 by RFIC designers from Texas Instruments and Qualcomm. The company's radar contains a 28-nanometer chipset and uses machine learning algorithms to reconstructs images from what the system senses. The company is based in Bangalore, India, and it is targeting radars that operate over the 79 GHz band and can support different modulation schemes, including FMCW used in short range radars.
Zendar. This company raised $4.3 million to develop autonomous car radars last year after passing through Y Combinator, the Silicon Valley business incubator. Zendar was founded by Jimmy Wang – previously a research scientist for Here Technologies, which sells digital road maps drafted using lidar – and former Nokia researcher Vinayak Nagpal. That is all we know right now.
Lunewave. Lunewave takes advantage of 3D printing to lower the cost of its high resolution radar. The company’s antennas are based on Luneburg lenses – orbs about the size of a quarter that can be manufactured to electronically steer specific bandwidths and frequencies. Lunewave is still in the research and development phase, but it claims that the spherical antenna can scan 360 degrees out to 300 meters. Hao Xin, an electrical engineering professor at the University of Arizona, was the lead inventor of the technology.