Tesla’s Autopilot Team & Cybertruck Visit New Orleans For CVPR
Tesla’s Autopilot team recently visited New Orleans for the Conference on Computer Vision and Pattern Recognition (CVPR). Ashok Elluswamy, Tesla’s Director of Autopilot Software, shared a photo of Tesla’s boot with an invitation for those in the area to stop by, talk with the engineers, and see the Cybertruck.
The Tesla Autopilot team is at CVPR this year! If you are in New Orleans, do stop by the booth, talk to the engineers & checkout the Cyber. pic.twitter.com/ni3qpOUaKa
— Ashok Elluswamy (@aelluswamy) June 21, 2022
This year, CVPR describes itself a hybrid conference with both in-person and virtual attendance. It’s being held at the New Orleans Ernest N. Morial Convention Center, which is within walking distance along the scenic Mississippi River.
And a look at the floor plan shows that Tesla has one of the largest booths. Other companies attending the AI-focused event included TikTok, Meta, Argo AI, Apple, Cruise LLC, Zoox, LG and Microsoft.
Although Tesla is known for its electric cars, Elon Musk has repeatedly shown examples of Tesla’s future as an AI robotics company. During Tesla’s Q1 2021 earnings call, Elon Musk said that Tesla is developing one of the strongest hardware and software AI teams in the world. He went on to describe the evolution of Tesla’s technologies that were developed to solve the problem of self-driving. And remember, this was before Tesla revealed the Dojo Supercomputer at AI Day later in 2021. Elon Musk said:
“Although, like, right now, people think of Tesla as — a lot will think of Tesla as a car company or as an energy company — I think long term, people will think of Tesla as much as an AI robotics company as we are a car company or an energy company. I think we are developing one of the strongest hardware and software AI teams in the world. Certainly, we appear to be able to use things with full self-driving that others cannot.
“So, if you look at the evolution of what technologies we developed, we developed them in order to solve the problem of self-driving. So, we couldn’t find a powerful enough neural net that runs a computer, so we designed and built our own. The software out there was really quite primitive for this task.
“And so we built the team from scratch and have been developing what we think is probably the most advanced real-world AI in the world. And then it sort of makes sense that this is kind of what needs to happen, because the road system is designed for a neural net computer, our brain. Our brain is a neural net computer. And it’s — the entire road system is designed for vision with a neural net computer, which is because it’s designed for eyes and the brain.
“And so if you have a system which has very good eyes, you can see in all directions at once, you can see three focal points ahead or forward, but it never gets tired, it’s never sort of texting, it has redundancy, and its reaction time is superhuman, then it seems pretty obvious that such a system would achieve an extremely high level of safety, far in excess of the average person.
“So, that’s what we’re doing. Then Dojo is kind of the training part of that. So, because we’re — we have over 1 million cars, and perhaps next year we’ll have 2 million cars in active use, providing vast amounts of video training data that then needs to be digested by a very powerful training system. And currently, we use Tesla training software.
“So, we developed a lot of training software, a lot of labeling software, to do, to be able to do surround video labeling, which is quite tricky. This means all eight cameras simultaneously at 36 range a second per camera labeling video over time. There wasn’t any tool that existed for this, so we developed our own labeling tool. Then, taking it a step further, obviously, the Holy Grail is auto labeling.
“So, now we’re getting quite good at auto labeling, where we do — where the trainers train the training system and then the system auto labels the data and then the human laborers just need to look at the labeling to confirm that it is correct and perhaps make edits. And then every time an edit is made, that further trains the system. So it’s kind of like a flywheel that’s just sort of spinning up. And really, the only way to do this is with vast amounts of video data.
“So, then we need to train this efficiently. So Dojo is really a — it is a supercomputer optimized for neural net training. We think Dojo will be probably an order of magnitude more efficient on, say, not sure what the exact right metric is, but say, per frame of video. We think it will be an order of magnitude more cost-efficient in hardware and in energy usage for a frame of video compared to a GPU-based solution or compared to the next best solution that we’re aware of. So, then, possibly that could be used by others.
“It does seem as though over time — I mean, just an observation, I think basically is the fact that neural net based computing or AI-based computing is more and more of the compute stack. We — conventional computing — well, perhaps heuristics-based computing is still going to be important, still going to be very important, but it will become, but neural net will become a bigger and bigger portion of compute. So, a long story, but I think, yeah, probably others will want to use it too and we will make it available.”
Have a tip for CleanTechnica? Want to advertise? Want to suggest a guest for our CleanTech Talk podcast? Contact us here.
Latest CleanTechnica.TV Video
CleanTechnica uses affiliate links. See our policy here.
