Teslabot Optimus Gen 2

13 thoughts on “Teslabot Optimus Gen 2”

1. Hopefully, unlike SpaceX, it won’t move fast and break things.

2. If you want to impress me: have it juggling chainsaws while riding on a unicycle navigating through an obstacle course.

3. I learned from MIT Leg Lab/Boston Dynamics that bipedal walking is a free swinging pendulum motion.
   Optimus team has to learn about free arm and leg swing.
   If I were a physician I’d diagnosis this bot with cerebral palsy.

4. I won’t really be impressed that Optimus can really replace humans in a majority of labor roles until it can dance like Michael Jackson (including speed).

   While I’m glad Tesla is focused on making inexpensive robots, I do wonder if they over-optimized too soon. Maybe they should have started with a fully flexible and faster moving design and then worked on getting cost down?

   Or start at both ends – near-human-capability-but-expensive, vs. cost-optimized, and work their way inward over time to the best mix of cost-capability points to meet actual industry needs?

   • Someone on one of the videos about it noted that they thought the limits on speed looked like they were compute constraints rather than hardware constraints. The thing may be able to move a lot faster if it had the brainpower to control such movements and keep aware of its environment.

     It looks like its controls are (like the sorting behaviour of the earlier version) learned by video training a neural net rather than coded in. Coding in the egg handling demo would have been excruciating and it looks like it is influenced from human input when it folds unused fingers away. So technology improving multimodal neural networks would directly improve the performance of Optimus bots.

5. It feels like it’s just on the cusp of graduating from the classic robot shufflewalk, which is the real signpost of when it’s advanced to a serious capability for whole body motion.

   The hand motions looked decent too.

6. Hands starting to look fine which is probably most important to the use case.

   Walking still looks like a 90 years old man who crapped his pants.
   Maybe this is part of the solution to the Terminator problem. If it can’t move around well, it’s no Terminator.

   • it being less physically capable is actually one of Musk’s stated methods of reducing the terminator risk.

7. I mean, the idea of of building a humanoid i s quite ludicrous. Something that people who read too much science fiction at their youth. For industry you want a robot with the intelligence and sufficient handling degree o of freedom for the task. Looking like a human is not needed. Also, these things look too scary and you don’t want them at home.

   • I’ve thought that way in the past. But if you want to make one robot form that can do a lot of tasks, including replacing humans in many roles, at least roughly humanoid is probably the shape to go with.

     E.g. during the robots-replace-humans revolution, we’ll still have lots of non-automated construction equipment that we won’t want to replace immediately. You can either design a robot to plug into each different machine, or design one that fills the human-shaped hole all those machines already have. I doubt this new generation of Optimus is up to that yet, but that’s the direction.

   • While I have some sympathy for this argument, it depends on the industry. If you have something moving between a factory floor and and office or spending time in, say a china shop or around expensive equipment you probably don’t want something with a backside as long as Boston Dynamic’s Spot spinning around. You want something that can change direction without banging into something — maybe while partway through a door if a situation changes and it has to reverse course. You don’t want something scuffing up your stairs with wheels or treads. You want something with about 2 arms (more is too expensive and complex, fewer is less functional) you’d like the arms to be on either side of the cameras which are probably forward facing and near the midline.

     There are roles for quadrupeds and wheeled bots and fixed or semi-fixed/slightly mobile platforms, as well as other forms but a humanoid form gives the potential, in theory, for more flexibility. What will really determines what use a humanoid robot is will not be the vision of Elon Musk and his supporters or his critics. The market will test these things over years and even decades and come to it’s own conclusion. Until now (or at least soon) the market had nothing like this to even test and experiment with.

8. Amazing.
   The speed of progress is astonishing. I would guess they will continue to refine it, and by the end of 2024, release the Gen 3, mass production hitting high speed in 2025, 50k apiece.
   The days of humans working 40 hours a week are coming to an end.
   I think AGI by the end of next year, put that in a Teslabot, and mass produce your labor force.
   Humanoid robots that are smarter and more capable then humans, and UBI for said humans, isn’t established by 2030, I’ll eat my hat.

   • While I am skeptical of the near term economic impact predicted (both because productivity technologies are always surprisingly slow to be adopted many years after becoming available and because human predictions often misunderstand how, when and where a technology will eventually be used when it gets to the market) I am also very impressed with the pace of advancement of this project. From something that couldn’t stand on it’s own to something that can walk and be trained via video input to the grace and fluidity we see today in such a short time is truly worth some admiration.

Comments are closed.