Zhou Ruifeng, a reporter from China Economic Weekly, | compiled
Recently, Tesla CEO Elon Musk said that the most important product this year and next will not be cars, but software that supports autonomous driving and robots.
For years, neither Tesla nor other car companies have been able to achieve the goal of fully autonomous vehicles. Some Tesla consumers have purchased $12,000 in self-driving packages, looking forward to the early arrival of fully autonomous driving technology.
Musk said: "I would be surprised if we can't achieve fully autonomous driving this year that is safer than humans." He predicts that fully autonomous driving technology will become "Tesla's most important source of profit." Musk made similar remarks a year ago. But Tesla's self-driving engineer at the time, CJ Moore, told U.S. regulators that Musk's remarks about self-driving technology "don't match engineering reality."
(Screenshot source: techxplore)
Tesla's autopilot system is controversial
Musk once said: "Anyone who relies on lidar is doomed to failure." ”
On the road to fully autonomous driving, Tesla has avoided other commonly used sensors such as lidar and bet all on cameras and artificial intelligence.
According to Tesla's official website, Musk pursued their system consisting of 8 "surround" cameras, which can collect data and import the car's "deep neural network".
But just like everything else related to Tesla, the system is controversial.
Recently, at the Consumer Electronics Show (CES) in Las Vegas, lidar manufacturer Luminar Technologies presented two cars, both rushing towards a small child model at a speed of 50 kilometers per hour. One of them, using Luminar lidar, stopped before encountering the mannequin, while the other Tesla crashed into the mannequin.
Luminar's Aaron Jefferson said the experiment was not validated by outside experts, and that it was intended only as a "side comparison" for people to "visually assess differences."
Jefferson says that in perfect driving conditions, with plenty of sunshine, cameras can do a lot of things. The problem is that in special cases, visibility can be affected by debris such as fog, light at sunset, or plastic bags.
Many companies in the autonomous driving space agree with Luminar and are focusing their development on merging cameras with technologies such as lidar.
But Tesla, under Musk's leadership, took a different approach. His approach is to feed real-time readings collected by cameras into an artificial intelligence system built around data collected by Tesla sensors over the years.
Kelian Weinberg, a computer scientist at Cornell University, said: "Tesla is betting on training algorithms by piling up large amounts of data."
According to Sam Abersamid of Marketplace Insights, self-driving systems are built around four functions: sensing the environment in real time, predicting what will happen next, planning the car's next move, and then executing.
"It turns out that this kind of prediction is much more difficult than anyone expected, especially in the face of pedestrians on the road," Abersamid said. The development of software built around cameras alone "has stagnated to some extent".
Tesla's self-driving technology, he said, is based on the hardware already in the car, and "they've pushed themselves into a desperate situation because they've sold the system to thousands of customers." ”
Also skeptical of Tesla's strategy is Jacques Aschenbloch, chief executive of French auto supplier Valeo. The company showcased its latest lidar technology at CES (International Consumer Electronics Exhibition).
Aschenblohi said: "Even if there is enough data, the camera is still not enough. He sees sensors as indispensable in urban areas or other places where vehicles and pedestrians are difficult to predict. Lidar is important for achieving more advanced autonomous driving.
Marco Bertonia, a professor at the University of Modena, said: "All sensors have advantages and disadvantages. The current state of technology is that there are still a lot of security issues with systems that rely solely on cameras. ”
(Screenshot source: Bloomberg)
More car companies prefer lidar
While Musk is skeptical about installing lidar sensors on cars, more and more competitors are quietly adding the technology to cars.
Lidar uses laser pulses to measure distance and create precise images of the surrounding environment. With the decline in costs and the increase in producers, lidar is being adopted by more and more car companies. At present, there are more than 150 lidar producers around the world, including many public companies, such as Velodyne, Luminar and Innoviz.
According to Bloomberg Financial News, as of January 2022, 17 car companies around the world have announced the launch of 21 models equipped with lidar, which are either in production or about to be launched. This number will continue to increase as lidar-based driver assistance systems, such as GM's Ultra Cruise, become available.
Although the matching software has not yet been developed, some car companies have installed lidar on their own cars. The idea is that the hardware is installed first, and the software can be upgraded at any time.
Recently, there has been continuous cooperation between car companies and lidar suppliers. In January 2022, Mercedes-Benz announced that it was working with Luminar and planned to integrate the technology into the cars that would be put into production soon. As early as 2021, Luminar had already reached a similar agreement with Volvo. BMW will also work with Innoviz to develop the iX model. Similarly, GM also chose Cepton as its lidar supplier.
Bloomberg said China is at the forefront of adopting new technologies. According to statistics, 70% of the current and upcoming models of lidar are from Chinese car companies, such as electric vehicle companies such as Weilai, Xiaopeng and Ideal.
(This article is compiled from Bloomberg, Reuters, techxplore)
Editor-in-charge: Yang Lin
(The copyright belongs to China Economic Weekly Magazine, and no media, website or individual may reprint, excerpt, link, repost or otherwise use it without authorization.) ）