Humanoid Robots for Moon and Mars

[sanman]

It now seems increasingly clear that Tesla is building the Tesla Bot to serve as a flexible type of automation for its Gigafactories, in order to minimize dependence upon human labor. Ideally, Tesla Bot would be flexibly re-tasked among different types of duties inside the plant like a human worker would.

[Oersted]

Latest footage of the Tesla bots:

- Musk says that he thinks they will more important for Tesla than cars.

I have no doubt that Elon envisions bots to play a very big role in Mars colonisation, since you can expect chronic work power shortages there for the first many years.

[Cheapchips]

The cute/staged nature of the investor day demo really makes it quite hard to judge what progress they've made since AI day. Their walk isn't really that dynamic next to RoMeLa's ARTEMIS or Apptronik's QDH, both of which reached their current state over a similar number of months.  With the other actions, grabbing etc, it's impossible to know if it's preprogramed or actually using something more general purpose.

Apptronik will publicly unveil their take on a general purpose humanoid robot later this year.  Their road to manufacturing is less clear than Tesla's but they are working with various other companies. With regard to space heritage, several of their team worked on NASA's Valkyrie robot and they have some sort of partnership with NASA for Apollo.  Frustratingly, I couldn't find proper details of the partnership's scope.

Wish I'd call the thread "General purpose humanoid robots for the Moon and Mars". 

(Couple of videos. Ignore the tethers.  With bipeds it's a safely feature when testing around people)



ARTEMIS is an academic robot, but this sort of walk cycle does make the Teslabot look a little bit geriatric.

https://twitter.com/DennisHongRobot/status/1621337645334003713?s=20

[HVM]

Apptronik is spin up from The Human Centered Robotics lab in The University of Texas, more than decade long experience in Robotics. Example Mercury Bipedal Robot is from 2018.

And same with Robotics & Mechanisms Laboratory. It's spin off from University of California's robot projects, also longer than decade of research in field...

I'm sure Tesla also hired people from "The Academy". But Optimus is not direct continuation of University project.

[Asteroza]

Optimus doing the Asimo shuffle footwork is a little disappointing, but understandable considering the foot used.

[edzieba]

Optimus doing the Asimo shuffle footwork is a little disappointing, but understandable considering the foot used.

Very few general-purpose (i.e. not single-purpose walkers designed around gait investigation) bipeds have managed to incorporate a passive-dynamic gait. Even Boston Dynamics' Atlas uses a full-active walk cycle.

[Cheapchips]

Video from the latest investor event.  Footage was gathered yesterday.   Looks like they're making progress, as much as you can tell from such things.

Comment on the grasping demo from Sr Software Engineer, Julian Ibarz.

Optimus performed its first end-to-end learned successful grasp today! You can see it below at 1X speed

This was learned from human demonstrations, no task specific programming was done. This means we can now scale quickly to many tasks.

https://twitter.com/julianibarz/status/1658580808880492544?s=20

[Cheapchips]

Changed topic title as there's a broad range of start ups building humanoid general purpose robots.  Assuming any of these efforts succeed on Earth, then they'll have applications for Moon / Mars bases.

Boston Dynamics have finally teased their electric version of Atlas, which will go into production at some point.  Boston are owned by Hyundai these days, so it'll have an interesting proving ground for usefulness.

BD are targeting something with 'superhuman' strength and agility, which is an interesting comparison to efforts from Tesla and Figure.

Interview with CEO Robert Playter:

https://spectrum.ieee.org/amp/atlas-humanoid-robot-ceo-interview-2667789605

Teaser video

[Cheapchips]

Figure showed off their 02 robot today.  They're hoping that the 2.25 kWh battery will deliver around 20 hours of 'useful work'.   Their robots are testing at some stations in a BMW factory, so I assuming 20 hours battery life is for that sort of activity. Not much walking but a fair amount of lifting and compute overhead.

The whole robot is 70kg. Food allocation for ISS crew is 1.831kg per day, for context.  2.25kW of electricity a day is essentially nothing in the context of a moon base.

There's still a lot riding on the various AI manipulation and reasoning continuing to develop at their current pace for these robots to be useful.

Hardware look to be pretty much good enough as of now, for in base use at least.

https://x.com/adcock_brett/status/1820792697315348640

[StraumliBlight]

Atlas is autonomously moving engine covers between supplier containers and a mobile sequencing dolly. The robot receives as input a list of bin locations to move parts between.

[InterestedEngineer]

Atlas is autonomously moving engine covers between supplier containers and a mobile sequencing dolly. The robot receives as input a list of bin locations to move parts between.

Atlas is doing a physical version of  JSON to YAML translation.  Yippee.

The program problem for whatever this relates to is not outputting YAML in the first place.  No robot needed, just a decent manufacturing engineer.

I'm here for all your software-analogy needs.  Very low rates.

[sanman]

Atlas is doing a physical version of  JSON to YAML translation.  Yippee.

The program problem for whatever this relates to is not outputting YAML in the first place.  No robot needed, just a decent manufacturing engineer.

I'm here for all your software-analogy needs.  Very low rates.

And yet the world is already built the way it is. This robot could be introduced into existing workplaces and handle them as they are, rather than these workspaces having to be re-engineered from the ground up.

Yes, if we use Musk's "Alien Dreadnought" analogy, the machine that builds machines can be engineered for high efficiency. But if you've got existing facilities that aren't operating at that level, then sometimes a patchwork solution may lower the barrier to get things operating.

PS: I like how Atlas v2 moves its hips independently of how the upper body is oriented. Although in this case, wheels might have worked even more efficiently & cheaply.

[JulesVerneATV]

not exactly looking like a person but more like a cross between a Robot MER and a Dog?

Watch: Wheeled robodog attacks extreme terrain at high speed
https://newatlas.com/robotics/lynx-wheeled-quadruped-prelaunch/
China's DEEP Robotics is about to add a new quadruped to its kennel called the Lynx. But rather than go for walkies on pads, the mid-sized pup rolls on four wheels and is built for all-terrain shenanigans – as you can see in the pre-launch promo.

[Twark_Main]

not exactly looking like a person but more like a cross between a Robot MER and a Dog?

Watch: Wheeled robodog attacks extreme terrain at high speed
https://newatlas.com/robotics/lynx-wheeled-quadruped-prelaunch/
China's DEEP Robotics is about to add a new quadruped to its kennel called the Lynx. But rather than go for walkies on pads, the mid-sized pup rolls on four wheels and is built for all-terrain shenanigans – as you can see in the pre-launch promo.

Movie killer robot: this thing.

Actual killer robot: DJI dropping Soviet-era antitank mine.

[sanman]

not exactly looking like a person but more like a cross between a Robot MER and a Dog?

Watch: Wheeled robodog attacks extreme terrain at high speed
https://newatlas.com/robotics/lynx-wheeled-quadruped-prelaunch/
China's DEEP Robotics is about to add a new quadruped to its kennel called the Lynx. But rather than go for walkies on pads, the mid-sized pup rolls on four wheels and is built for all-terrain shenanigans – as you can see in the pre-launch promo.

Movie killer robot: this thing.

Actual killer robot: DJI dropping Soviet-era antitank mine.

Could make for a frightful Black Mirror episode, though.
Or imagine this thing rolling/running around the Aitken basin.
In an off-world airless environment like Moon or Mars, hovering drone may not be feasible.
The wheely-pedal way may be the only practical way.
(Actually, is there a proper name for this kind of motion?)

[Cheapchips]

Musk replied to "thoughts on the first payload?" with "Cybertrucks and Optimus robots"

Looks like we can move this thread back out of Advanced Concepts in '2026'. 

https://twitter.com/elonmusk/status/1859078074303713447

[sanman]

Musk replied to "thoughts on the first payload?" with "Cybertrucks and Optimus robots"

Looks like we can move this thread back out of Advanced Concepts in '2026'. 

Heh, think of Cybertruck + Optimus as the natural progression or successors to Tesla roadster + Starman dummy.

 

[Cheapchips]

Figure posted an update a week or so ago. It's a less agile demo than Boston's latest, but it they're very proud of their 1000 placements a day on their BMW test stations. Something approaching real work. 

The somewhat nebulous improvement stats they flash up are relative to their last BMW video from three months ago.  In the context of space applications, the Optimus bots arriving at Mars do so with two years of software upgrades from the fleet back home.

Interestingly the speed isn't hardware limited here.  It's still the nascent control software that's stopping them doing the job as fast a human. 

The robots in the video are still their volume prototype. The mass production version is due to be revealed next year.

[sanman]

I still don't get why they need legs to move on a flat factory shop floor, when wheels would be simpler and more efficient.

[Guillerz]

While wheels are indeed efficient on flat surfaces, equipping robots with legs provides them with greater flexibility and adaptability. Legs allow robots to navigate a variety of terrains and obstacles that wheels might struggle with. This means they can be assigned to a wider range of tasks beyond just moving across a flat factory floor, making them more versatile in different work environments.