#1 most loved electric vehicle, solar, & battery news & analysis site in the world. Support our work today!


Autonomous Vehicles

Published on August 16th, 2020 | by Zachary Shahan

0

10 Tesla Autopilot, Full Self Driving, & Dojo Development Updates From Elon Musk

August 16th, 2020 by  


Tesla CEO recently took to Twitter to share a variety of updates on Tesla Autopilot and Full Self Driving. I don’t normally do list posts, but there were several interesting nuggets in this long tweetstorm and a list post seemed like the best way to roll through them in a clear way.

1. The Dojo Neural Net Training Center

At least since Tesla Autonomy Day in April 2019, Tesla has been working on a “Dojo” supercomputer to train neural networks (NN). At that time, he said, “We do have a major program at Tesla which we don’t have enough time to talk about today called Dojo. That’s a super powerful training computer. The goal of Dojo will be to be able to take in vast amounts of data and train at a video level and do unsupervised massive training of vast amounts of video with the Dojo program … Dojo computer.”

Elon brought it up again yesterday while making an appeal for AI and computer/chip experts to join the team that is developing Dojo. “Tesla is developing a NN training computer called Dojo to process truly vast amounts of video data. It’s a beast! Please consider joining our AI or computer/chip teams if this sounds interesting.”

He also provided several updates that will be covered further down this list.

2. Dramatically Better Autopilot, Limited Release in 2 Months

Ever since it was announced that Tesla was doing a massive rewrite to get around what seemed to be a wall in the development process, I and many others have been eager for the dramatic improvement implied. Current Autopilot/FSD is superb. I usually have it on. However, there are places where it is clearly not up to snuff and I have learned to watch out for them and take over at those times. The tease of a humongous improvement is enticing!

“The FSD improvement will come as a quantum leap, because it’s a fundamental architectural rewrite, not an incremental tweak. I drive the bleeding edge alpha build in my car personally. Almost at zero interventions between home & work. Limited public release in 6 to 10 weeks.”

3. Avoiding Potholes!

One of my biggest concerns about a potential FSD full-capability release has been that it still wouldn’t see and avoid potholes, which would make it a worse option compared to driving oneself in many areas. And the idea is that it should be better than driving oneself. So, perhaps most of all in this recent tweetstorm, I’m thrilled to see that the updated system is learning how to avoid potholes!

4. And Roundabouts! (Sort of …)

This is another top question I’ve had. It’s not been as big a question as potholes, since I knew it would be solved sooner or later. (And, apparently, the Volkswagen ID.3 can go through roundabouts.) However, it’s good to get confirmation of progress nonetheless. In response to a question about whether the next version of Autopilot will be able to do roundabouts, Elon wrote, “Not perfectly at first, but yes. Will take maybe a year or so to get really good at roundabouts worldwide. The world has a zillion weird corner cases.” Indeed — and there are many versions of roundabouts.

5. Wait, What is the Problem with Current Autopilot/FSD?

Trapped, You Say?!?!

Still with his mind on this topic earlier today, Elon tried to explain for us simpletons why Tesla had to do a massive rewrite of the Autopilot software. For fear of messing up his explanation, I’ll just quote him: “Autopilot was trapped in a local maximum, labeling single camera images uncorrelated in time. Now, it is not.” Here’s a definition of a local maximum. As for the rest, I think I understand, but I’m not confident enough in my interpretation to attempt to reword it. Feel free to add your own interpretation down in the comments.

6. Dojo will Enable Quicker NN Learning & “Operation Vacation”

Returning to Dojo, your question may be, what the heck is the purpose of Dojo? Viv explained in the following tweet and Elon responded “Exactly” to it:

Train faster. Master sooner.

“Whole Mars Catalog” adds a useful addendum:

As one final explanation and resource, Kristen provides a useful article that goes beyond tweet length to discuss the matter:

7. It’s an exaFLOP. For Realz.

What is an exaFLOP? Check here, here, here, and here for some background. The short answer is that it’s an insane supercomputer, the likes of which the world has not yet seen. Various top tech companies and the world’s largest economies are in a race to complete the first one. Count Tesla in here. Or, if it isn’t racing to be #1, the important point is that it’s joining the fun in order to try to train the AI for full self-driving capability and eventually robotaxis.

8. We are at Insemination, Not Birth

As the requests for engineers indicates, the Dojo supercomputer is not ready. Elon expects the first version of it to be ready in about a year, maybe. “Dojo is still being built. Maybe a year or so from V1.0.”

As I said, inception: ~0.01% capability.

9. Regulatory Approval? ¯\_(ツ)_/¯

One of the biggest questions is one Elon cannot answer. That question is: when will regulators approve the use of robotaxis?

“This is out of our hands, so hard to predict. Maybe on order of 10 billion km of real world driving will be enough to convince regulators,” Elon tweeted.

10 billion kilometers of real-world driving and what level of safety? How much safer will FSD have to be before regulators approve robotaxis? How will they judge/check interventions, potential accidents, accidents, etc.?

Theoretically, if they were twice as safe as humans, it would be an obvious decision to allow them. However, humans are often not fans of logic or rationality. Robotaxis will seem too scary for many. So, the level of safety will need to be dramatically better than that of human drivers. How much better? It seems that even Elon has no real idea. So, ¯\_(ツ)_/¯ it is.

10. Goat Honks. And Coconut Honks?

Okay, this one is not really about Autopilot or FSD, but it was in the same thread, so here it is. Elon confirmed that they are still working on making the horn honk like a goat. However, it will only be a feature in new cars.

I have to also emphasize a special request from our friend David Havasi, who worked at Tesla for 7 years. That request is: coconut sounds! A horn that sounded like coconuts would be so much cooler/nicer than a horn that sounded like a goat.

Okay, fine, that’s not an “Autopilot/FSD development update from Elon Musk” and my title is incorrect if I end here. The good news is that I also have a whole other article with fresh and exclusive info from Elon about the Autopilot team.

Tesla’s Team of ~300 Autopilot Jedi

Aside from all of the above, I recently got more insight from Elon on how many people work on the Autopilot team, and how that is split among hardware engineers, software engineers, and labelers. If you missed that exclusive article, check it out here: “Tesla Autopilot Innovation Comes From Team Of ~300 Jedi Engineers — Interview With Elon Musk.”

Related: Tesla Autostopping — 7 Notes On New Autopilot Feature

Want to buy a Tesla Model 3, Model Y, Model S, or Model X? Feel free to use my referral code to get some free Supercharging miles with your purchase: https://ts.la/zachary63404You can also get a $100 discount on Tesla solar with that code. No pressure.

Top photo by Kyle Field, CleanTechnica

 
 


 


Appreciate CleanTechnica’s originality? Consider becoming a CleanTechnica member, supporter, or ambassador — or a patron on Patreon.

Sign up for our free daily newsletter or weekly newsletter to never miss a story.

Have a tip for CleanTechnica, want to advertise, or want to suggest a guest for our CleanTech Talk podcast? Contact us here.


Latest Cleantech Talk Episode


Tags: , , , ,


About the Author

is tryin' to help society help itself one word at a time. He spends most of his time here on CleanTechnica as its director, chief editor, and CEO. Zach is recognized globally as an electric vehicle, solar energy, and energy storage expert. He has presented about cleantech at conferences in India, the UAE, Ukraine, Poland, Germany, the Netherlands, the USA, Canada, and Curaçao. Zach has long-term investments in NIO [NIO], Tesla [TSLA], and Xpeng [XPEV]. But he does not offer (explicitly or implicitly) investment advice of any sort.



Back to Top ↑