Levandowski announced Tuesday that his new company, ProntoAI, is developing a $5,000 driver assistance system for semi-trailer trucks, which will be responsible for automatic steering, acceleration and braking on highways.
To prove the feasibility of the new system, Levandovsky used the software to help his Toyota Prius drive across the country. Levandowski said that in October this year, the car traveled 4,987 kilometers from San Francisco to New York City. Apart from dealing with off-highway driving, such as refueling and rest, he has little control over the computer.
Photo: Anthony Levandowski, a former Google employee and Uber star engineer, returns to the field of driverless driving with plans for autonomous trucks. His new plan relies more on camera and machine learning technology than lidar and detailed maps.
In an official ProntoAI post, Levandovsky wrote, Yes, Im back!
Levandovsky has remained silent since Waymo sued Uber in February 2017. Levandowski was one of the founding members of Googles driverless team (later changed to Waymo). He left the company in January 2016 to start Otto, a driverless truck company. A few months later, Uber bought Otto and put Levandovsky in charge of the companys driverless business.
In the lawsuit, Waymo accused Levandowski of stealing thousands of technical documents from his former employer during his resignation, including the design of a special type of laser sensor, and using these documents to speed up Ubers research on drone technology.
Although Levandovsky was not listed as a defendant, when he was dismissed, he invoked the rights of the Fifth Amendment to the Constitution and opposed self-incrimination. Uber fired him in May 2017 and the two companies settled in February 2018.
As ProntoAI is exposed to the spotlight, Levandorfsky will bid farewell to Waymos moonshot project, which aims to provide a fully automated driverless system that never requires human participation.
Like Teslas Autopilot, ProntoAIs Coplot relies on people behind the steering wheel to monitor driving and take over cars when necessary. To ensure that people remain vigilant, the system uses an internal camera, which alerts people if they look off the road.
According to the Guardian, Levandorski has made several cross-country attempts at Coplot, including a test in Utah that ended with a system out of control. Every time, Levandovsky returned to San Francisco to start over. During the reporters test drive, Levandovsky once took over the system to deal with the thorny parallel problems.
Nevertheless, Levandovsky insists that his approach will make it possible for robotic cars to drive automatically on highways. He wrote: Simply put, the driverless industry has made two key mistakes: first, it has always been committed to the dream of fully automated driverless driving, and wants to be directly separated from all manual operations. Secondly, the industry is trying to pursue this false dream with crutch technology.
Crutch technology refers to detailed maps and lidars, a special laser sensor that most driverless car developers believe is crucial to helping cars identify obstacles and other road subjects.
Levandovsky was an early advocate of the technology and the first salesman for the Velodyne lidar system created by Dave Hall in 2005. Many lidar companies have improved the capability of this sensor over the past decade, but it is still very expensive and can only work within a range of about 250 meters.
ProntoAIs system does not use lidar, but uses six cameras to observe the road surface, and uses two neural networks to analyze images and make driving decisions. Cameras can see much farther and have much higher resolution than lidars, but very clever software is needed to interpret the large number of two-dimensional pixels they provide.
It is much easier to understand the feedback of lidar. But with the latest advances in machine learning and tensor processing hardware, the technology of dealing with cameras is improving. In his post, Levandovsky said it was a different and ultimately more promising way to solve the challenge of drone driving.
Levandovsky is not the only one who believes that a camera can outperform a laser. Elon Musk insists that his Tesla will one day use only cameras and radars to achieve full-automatic driverless driving.
Although TuSimple, an unmanned truck startup, does use lidar, most of its sensing and path planning systems are based on data provided by cameras. Wired magazine reporters recent experience in a TuSimple truck shows that it is feasible to use a system based on more cameras to achieve driverless driving on freeways.
For ProntoAI, the bigger problem now is how much more work, time and money will be needed before Levandovsky and his team can improve their systems to promote commercial deployment, rather than just sending people to monitor a moving truck on the highway.
This raises a bigger problem: Levandovskys reputation may affect ProntoAIs introduction of capital and talent, which hinders its continued improvement. His business theft has been a bad start. Moreover, in October, the New Yorker magazine reported that Levandovsky had ignored Googles rules and allowed vehicles to drive on unauthorized roads. In one case, a car accident happened, resulting in injuries to the spine of colleagues.
Whats more, Levandovsky claimed in an interview with the magazine that although ProntoAI puts comfort and safety first, security is not your top concern if your job is to promote technological progress. In addition to his negative news, Levandovsky also has many historical achievements that exceed peoples expectations. He drove an unmanned motorcycle to the Grand Challenges sponsored by the Defense Departments Advanced Research Projects Agency (Darpa), which was seen as the pioneering moment of the unmanned world.
Before Google had formed a driverless team, Levandovsky had let the driverless Prius cross the San Francisco Bay Bridge. He has a reputation for taking risks and taking shortcuts, and is also considered a charming and inspiring non-traditional thinker. These are precious characteristics of Silicon Valley. Moreover, a modified approach is not necessarily a bad thing for a problem that has proven to be more difficult than expected. (Selected from: WIRED Author: Alex Davies Compiler: Netease Intelligent Participation: Small) https://www.wired.com/story/anthony-levandowsk-pronto-self-driving-truck/Source: Netease Intelligent Responsibility Editor: Yao Dili_NBJS7522
Before Google had formed a driverless team, Levandovsky had let the driverless Prius cross the San Francisco Bay Bridge. He has a reputation for taking risks and taking shortcuts, and is also considered a charming and inspiring non-traditional thinker. These are precious characteristics of Silicon Valley. Moreover, a modified approach is not necessarily a bad thing for a problem that has proven to be more difficult than expected. (Selected from: WIRED Author: Alex Davies Compiler: Netease Intelligent Participation: Small)