While some technology companies and automakers have promised fully autonomous vehicles in the near future, most have acknowledged the complexities of developing hardware and software, as well as ensuring public safety. Innoviz Technologies Ltd. this month announced its InnovizTwo solid-state lidar system, which it said can help companies bring driver-assistance features to market sooner.
Rosh Ha’Ayin, Israel-based Innoviz said its InnovizOne is an automotive-grade, mass-producible package including a lidar sensor and software for machine learning-based detection, classification, and tracking. In February, the company announced that it was supporting Shaanxi Heavy Duty Automobile Co.’s autonomous truck project in one of China’s largest ports.
“When we started four and a half years ago, everyone said autonomous vehicles would be ready by 2021, but they’re still too expensive, and it’s difficult to validate,” said Omer Keilaf, co-founder and CEO of Innoviz. “Right now, the MVP is Level 3 autonomy, in which the car drives itself and the automaker takes responsibility, but only on certain roads.”
InnovizTwo intended to help validate safety affordably
“The major automakers won’t launch Level 3 until they feel it’s safe, and they need 100 million km [62 million mi.] to validate for safety,” Keilaf told The Robot Report. “Not that many carmakers have the technical ability themselves or are willing to be on that path.”
“We think providing lidar and computer vision for Level 2+ is the biggest opportunity for the next 10 years,” he said. “Volumes are not high yet, but they’ll start to grow in 2023 or 2024. Most carmakers are looking for something simpler, and in Level 2+, the car drives itself, but the driver must still pay attention for safety and liability.”
“Previous lidar systems cost $1,000, but InnovizTwo has a 70% cost reduction from InnovizOne, a higher resolution, and a wider field of view at low range,” explained Keilaf. “Carmakers can now develop Level 3 autonomy more safely and quickly, with validation crowdsourced through customers, somewhat like Tesla has done.”
“Carmakers can measure engagements on their own in real-life environments to assess risk,” he added. “With embedded, solid-state lidar, InnovizTwo is the right path for many carmakers that want to go to Level 3 but couldn’t. Other companies have used lidar with only four lines, poor resolution, which did not allow them to make the progress needed to collect data and validate the entire system.”
The new InnovizTwo lidar sensor features a compact design. Source: Innoviz Technologies
Looking ahead with Innoviz perception
By supporting Level 2+ autonomy, InnovizTwo provides an affordable path to Level 3 autonomy with software updates, according to Innoviz. It can complement other sensors, said Keilaf.
“InnovizTwo uses high-performance, proprietary architecture with industry-leading lidar perception,” he said. “Self-driving cars need to see 100 to 200 meters ahead with not just a camera. Radar is not able to identify objects, and lidar doesn’t stop working as only one of two cameras would [for stereoscopic vision]. Two cameras are not redundant to each other, and both are sensitive to light saturation. You can’t reach Levels 3 to 5 without redundancy, and Level 2+ today uses the human driver as redundancy.”
At Level 3+, autonomous vehicles may need to detect objects at 60 kph (37.2 mph) to 130 kph (80.7 mph), depending on the automaker, Keilaf noted. Higher speeds require longer range and higher resolution for identifying small objects, defined as one-third of wheel height, he said.
Partners for production
“We’ve worked very closely with BMW and Tier 1 partner Magna,” said Keilaf. “Sensors must deal with salt, ice, and water pressure in continuous and extensive auditing by BMW over the past three years. For sensors to replace human eyes, they need to meet a high bar of quality, and they need to meet automotive-grade certifications for manufacturing.”
“At high volumes, automakers don’t source from Tier 2 suppliers like us, which is why it’s important that we have relationships with four Tier 1 suppliers. Diversity is also important,” he said. “For Level 2 autonomy, the price point needs to be under $500. Eventually, we’ll see competition with lidars that can meet that point, but InnovizTwo is fast-forward toward Level 3 without changing hardware.”
The Innoviz team. Source: Innoviz Technologies
Samples of InnovizTwo will be available next year, said Innoviz, which raised $132 million in March 2019 and another $38 million in June 2019.
“Our volumes are still not high,” Keilaf acknowledged. “But by the time carmakers want to scale, InnovizTwo is available for their hardware refresh. It will make their time to market cheaper and faster.”
“We’re pursuing RFIs [requests for information] and RFQs [requests for quotation], and we’re also seeing some engagement with autonomous shuttles and robotaxis,” he said. “Not everyone needs to drive autonomously at 200 kph [124 mph] on the first day. It’s more important to get to market with a product that works, and InnovizTwo provides value for everyone.”
It’s well known that Musk is opposed to Lidar, but one of his main concerns seems to center on cost. Do you have any sense as to 1) what price might be acceptable and 2) does some combination-assuming a viable price, of Lidar and a combination of cameras and radar-work, that is, is there commercial viability down the road for such a combination?
Finally, are companies like NIO in China planning on using Lidar, and I assume, the Chinese being Chinese, that they have their own home-grown versions. Is that the case?