Tesla (TSLA) held its highly-anticipated “AI Day” on Thursday that was covered in a live blog by Benzinga. Here're some of the key takeaways from the technology-intensive presentation.
‘Tesla Bot' Humanoid Robot: CEO Elon Musk said that Tesla is building a humanoid robot, which will be friendly and take care of dangerous, repetitive tasks.
The 5’8” tall Tesla Bot can carry 45 lbs and will weigh 125 lbs. The maximum running speed of the robot is 5 mph, which as per Musk makes it safe.
The entrepreneur has in the past expressed concerns about artificial intelligence and thus the Bot is designed at a “mechanical level” in such a way that “you can run away from it, and most likely overpower it.”
Musk’s Tesla Bot uses some of the same software that drives its Autopilot driving system and the prototype is expected “sometime next year.”
Interestingly, the code name for the robot within Tesla is “Optimus” — a reference to Optimus Prime, the “Transformers” protagonist.
The Dojo Chip: Also on show at the AI Day was the custom chip dubbed “D1,” which would be a part of the company’s “Dojo” supercomputer.
The Chip uses a 7-nanometer manufacturing process and has 362 teraflops of processing power, as per Ganesh Venkataramanan, senior director of Autopilot Hardware, who gave the presentation on the silicon.
The automaker puts 25 of these D1 chips on a “training tile,” each of which has 9 PFLOP of processing power. 120 of these tiles can then be put together across multiple server cabinets to harness over an exaflop of power.
Venkataramanan said Dojo will be the fastest AI training computer. Other companies that make chips for AI training computers include Intel Corporation (INTC) and Nvidia Corporation (NVDA).
Musk said, “We should have Dojo operational next year” at the event.
Solving Computer Vision: A theme throughout the AI Day was Tesla’s emphasis on computer vision as an approach for solving full-self driving. Tesla AI lead Andrej Karpathy compared Tesla’s approach to vehicles with “effectively building a synthetic animal from the ground up.”
Tesla is building all the components of the ‘animal like' brain, the nervous system, and visual cortex, as per Karpathy.
As per Karpathy, Tesla is battling two main issues: temporary occlusions — akin to vision being blocked by moving vehicles — and road markings (for example lane merging signs), which appear earlier while the vehicle is on the road and remembering such signs.
In order to solve these issues, Tesla uses a spatial recurring network video module, which keeps track of the road with the aid of a space-based and time-based queue.
Tesla showcased how it labels data and has come up with auto-labeling in certain instances using clips from its own fleet. The company also has a 1,000-strong manual labeling team.
Last month, Tesla launched its FSDBeta v9, leaving many testers impressed.
Hiring and Beyond: The overall emphasis of the AI Day was “hiring,” something Musk had disclosed beforehand.
“We basically want to encourage anyone who is interested in solving real-world AI problems at either the hardware or the software level to join Tesla, or consider joining Tesla,” said Musk.
Upping the interest quotient after the unveiling of the Tesla Bot, Musk urged talent to “Join our team and help build this!”
However, what also is clear after AI Day is that Telsa does not want be seen simply as a car company. As Musk said at the event, it is arguably the “world’s largest robotic company.”
Even so, hard questions remain on the safety of FSD and the course the automaker is taking for its future.
“I suspect Tesla's gonna have to soften their marketing language around FSD, which would have a near-term negative impact on uptake. Long term it doesn't matter because Tesla will get there and autonomy will be a required feature for all carmakers,” said Loup Ventures’ Gene Munster on Twitter.