(“Dojo” nervous system)
Beijing time on October 30th news, according to foreign media reports, since Tesla launched the Full Self-Driving (FSD) beta version last week, the video released by the lucky people who have obtained the test qualification confirms that Autopilot has improved significantly, which was once used by Autopilot. The task that 2.5 saw as a challenge is no longer difficult, and the secret weapon behind it all lies in Tesla’s upcoming “Dojo” supercomputer.
Tesla owners who have used the FSD for a while know that the Autopilot kit has limited functionality, and everyone who has ordered the FSD knows it’s a work in progress that will be transformed with the help of Tesla’s neural network. more mature. At the same time, the Tesla AI team is developing Dojo to process larger amounts of information and improve the vehicle’s understanding of its surroundings. This is Tesla’s neural network training program designed to start decomposing 4D data, not the “? 2.5D” data that Autopilot previously used.
Musk has detailed the need for a more sophisticated self-driving system on an earnings call in the second quarter of 2020:
“The big milestone that’s really happening right now is the transition from 2.5D to 4D in autonomous driving systems or in automotive AI. 2D is more like recognizing pictures that are strictly related in time but relatively isolated, while 4D is like video Can play continuously.” The biggest problem with previous versions of FSD and Autopilot was that there wasn’t enough information transferred through the picture. Understanding the changing environment around you requires time stamping and greater accuracy, and the key is to move from what Musk calls images or 2D to video, or 4D.
Musk pointed out, “The research we’ve been doing so far has actually been focused on 2D, and while 2D is well correlated in time, it’s nowhere near as good as a full 4D system, which is actually really good. It is difficult to express in words how useful a complete 4D system is. The 4D system has the ability to input an isolated picture in time instead of a video, and the system can also turn the picture into a surround video. Therefore, the 4D system is adopted The car looks like it will be a big improvement.”
The navigation around the island reflects the operating efficiency of the new system to a certain extent. Musk emphasized in a tweet on August 14 that although FSD may not be able to navigate the roundabout perfectly at first, it can indeed navigate the roundabout. The Tesla chief predicts that it could take a year or so to truly master roundabouts around the world. In fact, previous versions of Autopilot had difficulty passing roundabouts, making it difficult to navigate without human intervention. Previously, a Tesla enthusiast posted a video of the Model 3 failing to drive out of the roundabout in Autopilot mode, and the driver had to take over the driving task.
The Tesla FSD beta proves that improved understanding is a prerequisite for Tesla Autopilot to function more accurately. Another video posted by James Locke, a Model 3 owner who got the FSD beta, shows the Model 3 navigating out of the roundabout with relative ease. What impresses Locke most is that there is no need for human intervention in the process, and Autopilot completes the driving task independently.
Dojo is about to be released with an updated FSD beta that will prove to be the answer to all of Tesla’s previous questions. With this newer and more sophisticated system capable of acquiring more information about terrain, surroundings and obstacles, Autopilot will be more accurate than ever before. With the regular release of new videos from the FSD beta, one will also find that Autopilot’s capabilities are improving every day.