SGS Approved Manufacturer of metal stamping parts
Recently, Tesla provided testers with the FSD Beta 10.12.2 beta software. In a recent tweet, CEO Elon Musk said that the update had been pushed to about 100,000 car owners, and revealed the 10.13 version. some information. (The full name of FSD is Full Self-Driving, which is “Full Self-Driving” in Chinese) SGS Approved Manufacturer of metal stamping parts
It is reported that the FSD 10.12.2 version is significantly smoother overall than the old version, and the steering wheel is significantly less jagged when making long turns, which means a smoother and easier ride for the occupants. The decision-making framework has also been improved in this release, improving the way the vehicle squirms forward to gain visibility, better predicting lanes, seeing pedestrians more accurately, and bicycles and motorcycles better, Tesla said. to determine whether the car is parked.
And a more important improvement is at hand, namely version 10.13, which Musk said will be the first version to support driving on roads without map data. The vehicle will rely entirely on its vision-based self-driving component to navigate roads that lack map information. SGS Approved Manufacturer of metal stamping parts
That’s a pretty big deal for FSD, Musk said, adding that the release is still months away. Additionally, version 10.13 will also make long left-hand turns better than the current FSD version.
It is understood that Tesla also plans to expand FSD to Europe this summer and gain a sizable group of owners.
Tesla is making a world without traffic accidents and congestion a reality through the most realistic and widely adaptable autonomous driving technology that is close to mass production. Tesla’s Autopilot is constantly evolving, and its chips have gone through 3 iterations.
Why did Tesla give up third-party chips and choose self-developed chips?
In the research report, CITIC Securities pointed out that in July 2016, due to a safety accident, Tesla stopped cooperating with Mobileye, and the vehicle used NVIDIA’s self-driving chip. The main reason is that on the one hand, it is dissatisfied with the slow progress of Mobileye, and on the other hand, it is dissatisfied with Mobileye’s use of Tesla owner driving data to improve the chip algorithm. In December 2017, Musk revealed to the outside world that Tesla is developing an autopilot chip, mainly because NVIDIA’s autopilot chip solution mainly relies on GPUs and consumes a lot of power. SGS Approved Manufacturer of metal stamping parts
Tesla Autopilot 3.0 hardware is 21 times more powerful than the 2.5 version chip. The huge increase in computing power is aimed at processing massive amounts of image information. The reason why Tesla’s automatic driving assistance system can quickly iterate and improve the experience is that it can pass through various uncommon but still possible “corner cases”. Deep learning (neural network) enables the system to handle more and more driving scenarios.
In short, deep learning is a learning method that simulates the learning mechanism of the human brain neural network system through hardware, which is similar to the process of people becoming familiar with and learning about external things. By continuously inputting data, Tesla manually or automatically annotates the correct “answers”, so that it constantly “evolves” itself, so as to quickly provide recognition rates and continuously cover all driving scenarios. SGS Approved Manufacturer of metal stamping parts
Whenever the vehicle encounters various “corner cases”, that is, encounters some “tricky” driving scenarios, or when the driver’s operation is inconsistent with the “expected” operation of the system, the vehicle will desensitize and anonymously report the actual situation. Upload to Tesla cloud server, and optimize the system through deep learning through huge centralized computing power. Of course, the system also provides users with the option of not uploading data. All data uploaded to the cloud is desensitized and anonymized.
Compared with companies that rely more on fleet testing, the advantage of Tesla’s FSD learning process is that most of the data is collected by vehicles that are actually driving on the road, and only a small part comes from the test fleet, which is closer to the daily driving scenario of the car owner, and this data is growing exponentially. SGS Approved Manufacturer of metal stamping parts
In order to process the huge amount of driving data, Tesla will increase the research and development of self-supervised learning technology to an absolute priority. Tesla has launched a major project code-named Dojo. It is a super-powerful training computer. The processed data is no longer at the picture level, but for video data, which can achieve algorithm performance at a lower cost. exponential improvement. SGS Approved Manufacturer of metal stamping parts
