Showcasing the latest progress in humanoid robots, Tesla’s road to “AI empire”

According to Automotive News, Tesla Robot’s official X account Tesla Optimus Prime recently updated an introductory video. This video showcases new advances in the control and execution capabilities of Tesla’s humanoid robot Optimus Prime. Optimus Prime can now classify objects by sight and even perform yoga moves

Are humanoid robots a hot trend?

Tesla stated that Optimus neural network training is completely “end-to-end”, that is, after inputting the video, it outputs the control signal. This is similar to the neural network training during the development of Tesla's self-driving FSD V12: processing all input signals and then outputting driving decisions. The intermediate process is fully handled by the neural network.

"Automotive News" pointed out that Optimus has many highlights in terms of perception, brain, movement and control: In terms of perception, through visual perception and joint position encoders, Optimus can automatically calibrate limbs and accurately locate the spatial position of limbs. . In terms of the brain, with pure visual technology and a fully locally deployed neural network, Optimus can quickly adapt to the environment and complete multiple tasks. It sorts blue and green building blocks into trays of corresponding colors. Even if someone disrupts the blocks while grabbing the blocks, Optimus can immediately adjust and adapt to the new environment and continue sorting the blocks. At the same time, Optimus can also straighten overturned blocks and perform new tasks such as "disrupting sorted blocks." In terms of motion control capabilities, Optimus can accurately grasp objects. When making movements, the robot's limbs, trunk, and finger movements are extremely flexible and close to humans. In addition, Optimus can also perform multiple single-leg support sports stretching actions, and can maintain trunk balance while stretching.

"Forbes" believes that humanoid robots are a very clear next trend and one of the main application carriers of artificial intelligence technology. For Tesla, cars are only a small part of the robot project, and the real huge application lies in humanoid robots, ranging from simple industrial fields to daily life and medical fields. But currently, Tesla faces two major challenges: first, the technological breakthrough of the physical body of humanoid robots, because the current humanoid robots do not have the ability to truly replace human occupations because the physical bodies are not refined enough; The robot's super brain, that is, the artificial intelligence part, is also a real challenge facing Tesla.

Dojo is the cornerstone

Currently, Tesla FSD and the underlying modules of humanoid robots have achieved a certain degree of algorithm reuse. FSD's algorithm mainly relies on neural network and computer vision technology, both of which will play an important role in the iteration of robot perception, decision-making and control technology. The autonomous driving and humanoid robot business is expected to have a strong synergistic effect.

With a powerful algorithm, it also needs strong computing power as support. The Dojo that Tesla mass-produced not long ago is the "cornerstone" Tesla has prepared for its AI empire. Dojo is a supercomputer developed by Tesla. It can use massive video data to complete "unsupervised" data annotation and training. It is a computer developed by Tesla to improve the performance of artificial intelligence products such as autonomous driving and intelligent robots. A powerful computing platform.

Literally, Dojo means "dojo, martial arts hall", which is also consistent with its meaning in the field of Tesla AI - a training place built for AI. On AI Day in 2021, Tesla released the Dojo supercomputer, but it was not yet complete at the time, with only the first chip and training module

The Detroit Times pointed out that Dojo’s final implementation unit is a supercomputing cluster called ExaPOD, which integrates 3,000 D1 chips based on 7nm process technology, including 120 training tiles, and can ultimately achieve up to 1.1EFlops (10 billion floating-point operations per second) peak computing power. For comparison, the computing power of a 3090Ti graphics card is about 40TFlops (1,000 billion floating point computing power per second), a difference of 6 orders of magnitude.

It should be noted that ExaPOD, as a supercomputing cluster, is not the final form of this supercomputer. Theoretically, it can be quantitatively expanded according to Tesla's requirements for greater computing needs to provide higher AI computing performance

In fact, according to the computing power development plan released by Tesla in June this year, Dojo will become the top five computing power facilities in the world in the first quarter of next year and will reach 100EFlops of super computing power by October next year. .

Now, the Dojo supercomputer has been put into production and is expected to become one of the five most advanced computers in the world by 2024. When running Occupancy Networks, a neural network model that tests automatic annotation algorithms and predicts the space occupancy of all objects around the car, Dojo can double its performance compared to NVIDIA A100.

Morgan Stanley analysts said in a report released on September 10 that Tesla’s Dojo supercomputer may promote the popularity of Robotaxi and its software services, causing Tesla’s market value to jump by nearly 6,000 One hundred million U.S. dollars. According to Reuters

Morgan Stanley analysts pointed out in the report that Tesla’s Dojo technology is expected to open up new potential markets, beyond the scope of the traditional car sales model. They said: "If Dojo can help cars 'see' and 'react', what markets can it open up? Imagine that any device equipped with a camera can make real-time decisions based on its field of view."

Morgan Stanley raised its revenue forecast for Tesla’s network services business to $335 billion in 2040 from the previous forecast of $157 billion. Jonas expects the unit to account for more than 60% of Tesla's core earnings by 2040. Tesla's expected price-to-earnings ratio for the next 12 months is 57.9 times, much higher than the 6.31 times of traditional American automakers Ford and 4.56 times of General Motors

The above is the detailed content of Showcasing the latest progress in humanoid robots, Tesla’s road to “AI empire”. For more information, please follow other related articles on the PHP Chinese website!