SAN FRANCISCO -- A startup based here is working to eliminate costly lidar sensors from the self-driving car equation.
Computer vision company Lvl5 launched Wednesday, and has developed software to construct three-dimensional maps from data recorded by camera sensors. The five-person startup has raised $2 million in seed funding, and plans to expand to 100 employees by the end of 2018.
"Accurate, up-to-date high-definition maps are the technological impediment holding back the autonomous vehicle industry," said Andrew Kouri, co-founder and CEO of Lvl5. "We wanted to make sure that there was a solution out there for all these companies saying they're shipping self-driving cars in 2020."
Kouri and fellow co-founder Erik Reed formerly worked as engineers at Tesla Inc., developing the first generation of the automaker's semiautonomous Autopilot system. Similar to Lvl5's approach, Autopilot relies on camera and radar sensor data to perceive the environment surrounding the vehicle.
Typically, automated driving systems rely on lidar sensors to create the three-dimensional images Lvl5 said it can construct with its software. However, such sensors, which resemble spinning coffee cans on top of self-driving test vehicles, are expensive, currently costing around $75,000 to $85,000 per sensor.
But the cost of lidar is starting to come down. Earlier this year, Waymo announced it had cut 90 percent of the costs out of its lidar system. And supplier Quanergy says its $250 lidar will be on the market later this year.
Kouri said Lvl5 is sidestepping this issue by developing an algorithm that can process two-dimensional camera data into a three-dimensional map. The startup has been collecting camera data through fleet drivers, including truck drivers and Uber and Lyft drivers. The data collection is conducted through an app, which drivers download and run while the phone is mounted inside the windshield, recording miles driven. Drivers are paid up to 5 cents per mile recorded. Lvl5 has mapped about 90 percent of U.S. highways via fleet driver crowdsourcing.
"In the long-run, we can install cameras into new vehicles, or hook into existing cameras in the vehicle," Kouri said. "The algorithm works the same."
Lvl5 is joining a competitive field of suppliers looking to map the world for self-driving cars. Legacy mapmakers Here and TomTom have been building high-definition maps through fleet vehicles equipped with powerful sensors. Startup Deepmap, which was founded by Google Maps veterans, plans to use lidar, camera and radar sensors on passenger cars with advanced driver assistance systems, and eventually fully self-driving cars, to build its three-dimensional map.
Kouri said the startup has been running pilots of its software with automakers, but will need a partnership agreement with a large Tier 1 supplier or manufacturer to enhance its ability to collect real-time data. And by only needing to install cameras and Lvl5's software, he claims the platform can be used in vehicles immediately.
"To make a self-driving car mass-produced, we no longer have to wait for lidar to get to a state where it can be mass-produced," Kouri said.