Anthony Levandowski, the autonomous driving technology engineer named as the culprit in a trade-secret suit against Uber by Waymo, has launched a new company called Pronto AI.
The company will begin delivering Copilot, a $4,999 driver-assistance-system kit for long-haul trucks, in mid-2019.
It is a camera-based software suite that can handle braking, throttling and steering for long-haul trucks when they are running on the highway.
While the system sounds promising, “the highway part is the ‘easy’ part” of autonomous driving, said Antti Lindstrom, a trucking analyst with IHS Markit.
Off-highway driving, “where there is other traffic and unpredictable behavior from other vehicles plus human beings, animals and the like, combined with weather conditions and defective signage, is where the going gets tough. I’m curious to see how this evolves,” Lindstrom told Trucks.com.
Levandowski was an original member of the Google self-driving unit that became Waymo. He left the company to start Otto, then joined Uber when it acquired Otto in 2017.
He was not sued individually when Waymo claimed in its February 2017 complaint that Uber had stolen its trade secrets. But the lawsuit named Levandowski as the individual alleged to have stolen technical files when he worked at Google – files that later helped Uber accelerate its research and development.
The suit was settled early this year, a few months after Uber and Levandowski parted ways.
Levandowski said the Pronto system – which he tested with a hands-off, 3,100-mile cross-country drive of his own Toyota Prius sedan in October – uses predictive software paired with an array of six cameras.
The cameras send images of the road to a pair of computerized “neural networks” that analyze the images and makes instantaneous decisions regarding steering, acceleration and braking.
Neural networks function much like the human brain, learning from the information they take in and continually updating their responses in reaction to new information.
Levandowski says the system is faster, more accurate and more responsive than the radar and lidar-based technology on which most autonomous driving systems are based.
The controversial engineer was an early proponent of lidar – a laser-based system for detecting obstacles, pedestrians and other vehicles on the road – but now calls it a “crutch technology.”
The Copilot system is a Level 2 autonomous system, meaning it doesn’t function in all instances and requires a human driver to remain fully aware and capable of operating the vehicle at any time. It is much like Tesla’s often misunderstood Autopilot system.
A spokeswoman for Levandowski said he was not available for comment.
In a post on Medium.com, he said that he does not believe fully autonomous vehicles will arrive “for many more years.” Today’s autonomous driving software “isn’t good enough” to match the instincts of human drivers, Levandowski said.
Learning to drive
Copilot isn’t a technology that tells vehicles how to drive, but technology “that can learn to drive the way people do,” he wrote in the post.
Delivering a commercially viable driver-assistance system is the first step in achieving safe autonomous driving, he said, and that is Copilot’s aim.
Levandowski acknowledged his sometime precarious position in the autonomous driving world, commenting that he has been “painted into a villainous caricature” and landed in “hot water” because of “my refusal to toe the corporate line.” He said he has a “well-deserved reputation for brash optimism.”
Pronto AI was founded earlier this year by Levandowski and Ognen Stojanovski, a corporate attorney and Stanford University researcher who headed government relations at Otto, the self-driving truck company Levandowski and others started in 2016.
Levandowski “certainly is one of the pioneers and most experienced people in autonomous vehicles, hence this is likely to be good technology,” Egil Juliussen, autonomous vehicles analyst at IHS Markit, told Trucks.com.
“But it is too early to say if it is a breakthrough advance,” Juliussen said.