Tesla’s autopilot team recently visited New Orleans for a conference on computer vision and pattern recognition (CVPR). Ashok Elluswamy, director of Tesla’s autopilot software, shared a photo of Tesla’s boots, stopped by locals to talk to engineers and get a look at Cybertruck.
The Tesla Autopilot Team is participating in CVPR this year! If you’re in New Orleans, stop by the booth, talk to engineers, and check out Cyber. pic.twitter.com/ni3qpOUaKa
— Ashok Elluswamy (@aelluswamy) June 21, 2022
This year, CVPR describes a hybrid conference with both face-to-face and virtual participation. It is held at the New Orleans Ernest N. Morial Convention Center, which is within walking distance along the scenic Mississippi River.
Looking at the floor plan, you can see that Tesla has one of the largest booths. Other companies that attended the AI-focused event included TikTok, Meta, Argo AI, Apple, Cruise LLC, Zoox, LG, and Microsoft.
While Tesla is known for electric vehicles, Elon Musk has repeatedly shown examples of Tesla’s future as an AI robot company. In Tesla’s first-quarter 2021 earnings announcement, Elon Musk said Tesla is developing one of the most powerful hardware and software AI teams in the world. He further described the evolution of Tesla’s technology developed to solve the problem of autonomous driving. And remember, this was before Tesla unveiled the Dojo supercomputer at AI Day in late 2021. Elon Musk said:
“I think many people think of Tesla as an automobile company or energy company, as it is now, but in the long run, Tesla will think of it as an AI robot company like us. Automobile company or energy. Company. I think we are developing one of the most powerful hardware and software AI teams in the world. Indeed, to be able to use things with fully automated driving that no one else can. is.
“So, looking at the evolution of the technology we developed, we developed them to solve the problem of autonomous driving, so we couldn’t find a neural net powerful enough to run a computer. So I designed and built it myself. The software there was very primitive for this task.
“So we’ve built our team from scratch and developed what we think is probably the most advanced real-world AI in the world. And the road system is a neural net computer, or us. Designed for the brain, it makes sense in a way that this is one of the things that needs to happen. Our brain is a neural net computer, and it’s — a road system. The whole thing is designed for vision with a neural net computer, because it is designed for the eyes and brain.
“Therefore, if you are using a system with very good eyes, you can see all directions at once and see the three focal points forward or forward, but never get bored, It’s not like a text message, it’s verbose, its reaction time is superhuman, and it’s pretty clear that such a system achieves a very high level of security that far exceeds the average person. Seems to be.
“That’s what we’re doing. And the dojo is part of that training. That is, we own more than a million cars and probably have two million cars active next year. And will provide a large amount of video training data that needs to be digested with a very powerful training system. And now we are using Tesla training software.
“Therefore, we developed a lot of training software, a lot of labeling software to be able to perform surround video labeling, but this is a very difficult task. This is a time for all eight cameras. This means that it will occur simultaneously in a range of 36 per second for each camera that labels the video over time. There was no tool for this, so we developed our own labeling tool. Going one step further, obviously, the Holy Grail is auto-labeling.
“Now I’m very good at auto-labeling. After the trainer trains the training system and the system auto-labels the data, human workers look at the labeling and it’s correct. You need to make sure that you fix it, you probably edit it, and each time you make an edit, the system gets more trained, it’s like a flywheel, it’s like spinning. And in fact, the only way to do this is to use a large amount of video data.
“Then we need to train this efficiently. So Dojo is really a — a supercomputer optimized for neural net training. Dojo doesn’t know what the exact metric is, for example. But for each frame of video, I think it’s probably an order of magnitude more efficient. Compared to GPU-based solutions, or compared to the next best solutions we recognize, with hardware. We consider it to be an order of magnitude more cost effective in terms of video frame energy usage, so perhaps it could be used by others.
“It looks over time. So it’s just an observation, but I think it’s basically the fact that neural net-based or AI-based computing is becoming more and more computing stacks. We — traditional computing—well, perhaps heuristic-based computing is still important and very important, and as it will be, neural nets will become an increasingly large part of computing. So, it’s a long story, but maybe others will want to use it, and we’ll make it available. “
Check out our brand new E-BikeGuide.. If you are interested in electric bikes, this is the perfect place to start your e-mobility journey!
Thank you for CleanTechnica’s originality and cleantech news coverage? Consider becoming a member, supporter, technician, ambassador, or Patreon patron of CleanTechnica.
