The Tesla 2022 AI Day event will be held in Palo Alto, California on September 30, 2022 North American time (expected October 1, Beijing time), where the Tesla Bot, which brings science fiction to reality, is expected to make its debut. In addition, the latest developments in Tesla’s self-driving technology and Dojo supercomputer may also be announced on that day.
Looking back at last year’s AI Day event on August 20, the launch of the Tesla Bot was a big, heavyweight egg. Tesla says it stands 1.72 meters tall, weighs 56.6 kg, has a screen on its face that displays information, and has human-level hands and powerful feedback sensing for balance and agile movements. In just one year, Tesla will bring this egg to life.
Not long ago, in an article published in China NetNewsWire magazine, Elon Musk wrote: “Tesla robots are initially positioned to replace people in repetitive, boring, and dangerous jobs. But the visionary goal is for them to serve thousands of households, such as cooking, mowing lawns, and caring for the elderly.”
“The Tesla robot is close to the height and weight of an adult, can carry or hand-carry heavy objects, can also walk in small steps, and the screen on its face is the interactive interface for communication with people,” Elon Musk said, “You may wonder why we designed this robot with legs? Because human society is based on the interaction of a bipedal humanoid with two arms and ten fingers. So if we want a robot to adapt to its environment and be able to do what humans do, he has to have roughly the same size, shape and capabilities as a human.”
According to Elon Musk, “Thereafter, the utility of humanoid robots will increase every year as production scales up and costs drop. In the future, a home robot could be cheaper than a car. Perhaps in less than a decade, people will be able to buy their parents a robot as a birthday present.”
To achieve ultra-high computing power for AI training while extending bandwidth, reducing latency and saving costs, Tesla built the Dojo supercomputer. the Dojo supercomputer’s individual training module consists of 25 of Tesla’s self-developed neural network training chips, the D1 chips. Since each D1 chip is seamlessly connected together with extremely low latency between adjacent chips, the training module maximizes bandwidth reservation, and with Tesla’s own high-bandwidth, low-latency connector; the arithmetic power is up to 9 PFLOPs (9 trillion times) and I/O bandwidth is up to 36 TB/s in a volume of less than 1 cubic foot.
Thanks to the training module’s ability to run independently and infinite linkage, the Dojo supercomputer is a “performance beast” with no theoretical upper limit on performance expansion. In practice, Tesla will assemble 120 training modules into ExaPOD, the world’s premier artificial intelligence training computer. Compared to other products in the industry, it offers a 4x performance increase at the same cost, a 1.3x performance increase at the same energy consumption, and a 5x space savings.
Since the release of the Dojo supercomputer last year, the development process of Tesla’s FSD (Full Self-Driving capability) has continued to accelerate. At this year’s AGM, Elon Musk revealed that FSD version 10.13 has been under study for some time and that Tesla has made “some very important architectural improvements,” such as improving the complexity of left turns. He said that Tesla is now widely deployed in North America FSD beta version, and will start a large-scale test of the fully autonomous driving capability (FSD) software this year, the beta version of the software has exceeded 40 million miles, and the end of the year is expected to exceed 100 million miles.
Elon Musk said at last year’s AI Day event, “What we’re trying to do at Tesla is make artificial intelligence that people like and that works …… and there’s no doubt that it’s going to be good artificial intelligence.” What surprises will Musk bring this year? Let’s wait and see.
