[ad_1]
Tesla presented its latest and greatest features in vehicle autonomy during its recent day of independent investors. The brain of the system is the Tesla Full Self-Driving Computer (FSD), which is now included in all Teslas products being produced. The Autopilot software suite includes 8 cameras, 12 ultrasonic sensors, a radar, a GPS, an inertial measurement unit and sensors that measure the angle of the steering wheel and the accelerator pedal.
- This article is provided by EVANNEX (which also manufactures Tesla accessories for the aftermarket). Author: Charles Morris. The opinions expressed in these articles are not necessarily ours at InsideEV.
 |
Above: Tesla's camera and Waymo lidar (Images: Tesla / Waymo)
LIDAR, a technology that uses pulsed laser light to measure distance from a target, is not an asset, and is preferred by most other companies working in autonomous driving (Uber, Waymo, Cruise). In his lecture on Investor Day for Autonomy, Elon Musk, iconoclast of forever, dismissed lidar in terms reminiscent of those he had used to describe hydrogen fuel cells. Lidar is "a crazy race," he said. It is "expensive" and "useless" and "anyone relying on lidar is doomed. It's like having a lot of expensive appendages. For example, an appendix is bad, and now you have a whole series. It's ridiculous, you'll see.
As reported by TechCrunch, lidar supporters cite its ability to better weather adverse conditions, such as rain, fog or dust, compared to cameras. The problem is that it is expensive and it consumes a lot of energy, which is why the Tesla system depends on the cameras. Andrej Karpathy, Senior Director of Artificial Intelligence, explained that visual recognition gives a better idea of the real world. He explained that lidar systems had trouble distinguishing things such as the difference between a plastic bag and a rubber tire. "In that sense, lidar is really a shortcut," said Karpathy. "This circumvents the fundamental problem, the important problem of visual recognition, necessary for autonomy. This gives a false impression of progress and is ultimately a crutch. "
With cameras, Tesla relies on the vast network of neurons containing actual driving information recorded by thousands of Tesla vehicles equipped with an autopilot on the road. Using various AI techniques, Tesla teaches his system to recognize and respond to the wide variety of situations that can be encountered in nature.
"Everybody trains the network all the time," said Musk. "Whether the autopilot is activated or not, the network is being trained. Every kilometer traveled for the car that is [equipped with Autopilot Hardware version 2 or above] is the formation of the network. "
& # 39;
Above: Anthony Levandowski, former Google / Waymo engineer, returns to lidar and says: "Elon is right" (Youtube: TechCrunch)
Some experts in autonomy agree with Musk. According to Gizmodo, Cornell researchers say cameras can compete with lidars if they are properly mounted. Lidar systems are designed to provide a 3D image of the vehicle environment and the road. That's why they often climb to the top of a vehicle for a better view. However, in an article that will be presented at the upcoming Computer Vision and Pattern Recognition Conference, the Cornell team explains that a pair of inexpensive cameras mounted behind the windshield of a vehicle can produce images stereoscopic data that can be converted to 3D data almost as accurately as possible. that generated by the lidar, at a fraction of the cost.
Mashable reports that the British start-up Wayve is part of the camp without the need for lidar. The company said it did not need multiple sources of data: a GPS system, cameras and a powerful computer were enough to teach cars to drive as well as humans.
Other players do not agree. AutoX, a Chinese autonomous driving company called AutoX, has deployed a camera-driven automatic steering system in 2017, but has now added lidar sensors to its vehicles to provide redundancy and additional contribution.
Musk is not totally opposed to lidar – SpaceX uses it in some applications. For vehicle range, however, he believes that Tesla's suite of sensors and its many actual driving data will be quite sufficient.
===
Written by: Charles Morris; Source: TechCrunch, Gizmodo, Mashable
- Editor's Note: EVANNEX, which also sells replacement equipment for Teslas, has kindly allowed us to share some of its content with our readers, free of charge. Our thanks go to EVANNEX. Check out the site here.
[ad_2]
Source link