@johnkoetsieri
iAccount based inCanada
About this account
- Account based in
- Canada
- Connected via
- Canada App Store
Account-level information from X, not a live location or the device used for a specific post.
What can Astra do when given a humanoid embodiment?
We built HomeBody to find out. Controlled by GPT Astra, it carries out long-horizon tasks in a previously unseen kitchen—from tidying up across the room to retrieving remembered objects from ambiguous requests—without environment-specific training data or additional policy learning.
Here's how we did it 👀: tml.stanford.edu/homebody/
John Koetsier retweeted
I'm having a Meta hangover.
Realizing that I was wrong to be excited at all about its new "VR Glasses."
I predict another sales disaster is coming.
And it isn't just because of all the people who keep telling me "I'm not wearing glasses."
I'm on a podcast with @SadlyItsBradley, @ASychov, and @Shodah10 talking about this and that made me realize that I am just wrong about being excited about Meta.
Which will be up this weekend. Bradley has long been in the VR/AR space and Artur and Shodah run a VR headset and metaverse company. So quite a bit of expertise.
They all are far more critical than I was with my first impressions. And they aren't alone. I've been talking to quite a few nerds in San Francisco and they just show no excitement about these devices.
And it honestly doesn't get close to the software experience, and content, of the Apple Vision Pro on a number of different levels.
The crappy avatars in Meta, when compared with Apple's, which are full AI/gaussian splats, is one signal that Meta just isn't aiming at the right thing.
I said "the Holodeck is here" but on a more critical look, no, they aren't even close.
To get us to the Holodeck we need far more impressive AI than Meta Muse has to offer us today. World Models, like what you are seeing @theworldlabs or @odysseyml or others like @pixverse showing off.
Now if they called them "the ultimate Meta Muse ecosystem/glasses" then I might feel differently. At least then we wouldn't be having fights about the field of view.
Because they called them "VR Glasses" instead of "Muse Glasses" VR fans rightfully called out the narrow field of view, which means immersion isn't nearly as interesting as on a bigger VR headset.
Calling them VR glasses is a huge mistake.
It means we all are comparing it to the Quest 3. But the Quest 3 doesn't have a cord to a puck and has real immersion, which these really don't.
But is is @alexandr_wang's post about its new holdable, or maybe wearable, AI device that tells me that vision at Meta is off: nitter.cf/alexandr_wang/status/2…
I will leave it to you to do the critical thinking to understand why that's such a bad way to introduce a new surveillance technology into people's lives.
Which is what this is.
Compare to how Google talks about use cases and utility. Yeah, Google isn't as sexy as having Raybans' brand, but at least Google has a vision for how it will help bring super powers to us and doesn't make us feel dirty about the direction it is going. In its glasses it showed someone getting help fixing a bicycle. Utility and use cases that give us super human powers will be a lot better way to get people into this new world.
Meta did a good job with its auto glasses, that can help people who are hard of hearing hear better. It should have had a bunch more use cases to show off why we need all these things.
It just all misses that there is a new consumer that is unsatisfied. Ones that use AI all day long to run their lives and businesses.
I could go on for hours (we did on the show which should be up by Sunday night), but I'll stop here.
I'm sorry for getting overly excited, which is what I do.
I try to be optimistic about new things, and it's clear that this company doesn't have good vision for where we are going over the next decade.
Which leaves a HUGE opening for not just the Chinese (ByteDance and Pico are working on devices that are far more AI-centric and far more open than these) but both startups (several are arriving over the next few months) and companies like OpenAI who will focus on what AI can do for you, rather than showing off a cute mascot shaking its ass at us.
And I'm a lot more excited about the AI that an American Company @getVITURE is bringing by the end of the year.
I should listen more to the feedback from you all who say you don't want these devices.
And I should have paid more attention to the fact that the software in these just isn't good enough to take us into an AI world, which is why they didn't let attendees at its own developer conference this week really get a good try of them.
I know that sometime in the next decade you'll probably change your mind, but not yet.
And the devices that really bring AI's promise of bringing us super powers aren't here yet, and this week didn't change that, so I can't recommend you buy these yet unless you are a developer or an extreme early adopter. Even then.
If I had listened, I wouldn't have a Meta hangover today.
John Koetsier retweeted
Good morning from @MojaveAirport! We're flying autonomously in complex, towered Southern CA airspace. Our Cessna 208B Caravan is demonstrating auto-taxi, auto-takeoff & auto-landing, all while our remote pilot at Reliable's HQ 300 miles away communicates with Air Traffic Control.
John Koetsier retweeted
The biggest thing between humanoid robots and real work isn't intelligence, or better hands, or a bigger battery.
It's a fence.
Digit 5 is being built to take it away, to notice people and yield to them, and work in the same space they do.
Framed well by @johnkoetsier in @Forbes: forbes.com/sites/johnkoetsie…
John Koetsier retweeted
We built a new kind of robot for the underdogs.
Today, robotics is stuck between ~$100k systems and low-cost robots not meant to survive deployment.
We think builders deserve better. Our mission is to develop the best hardware platform for others to deploy on.
Meet Feather, an affordable general-purpose robot designed for real work:
• 1m reach
• Human strength
• 10-hour battery
• Shipping for the past year + available today
Founded by @whoishoa and Parsa Bakhtiari, Feather is coming out of stealth with $7.6m in pre-seed funding led by @GradientVC, with participation from @BuilderVC, @geometryvc, @SEEDInnov, and Virgo VC.
We’re hiring across engineering, ops, and go-to-market. If you’re an absolute maverick, join our cause today – feather.dev
Google's testing data centers in space by sending up a satellite with 4 of TPU AI chips.
Biggest problem: cooling. Vacuum is an insulator.
Google's using passive dissipation, and it's a massive problem.
Radiators: 300W/m2
TPUs: 100,000 Wa/m2
At scale they need active cooling.
In today's Humanoid (almost) Daily:
- Tesla Gen 3 Optimus design leaks
- Japan is exploring military support roles for humanoids
- 80% of humanoid robots are for industrial use
- Saudi Arabia could deploy up to 10,000 robots over five years
Find me on LinkedIn to subscribe!
🤖 Made with AI
Where's the weight in a humanoid robot?
Mostly in actuators ... over half, according to a report by IDTechEx. Guess what: electric motors are heavy.
Most of the rest is battery and structure.
Other includes sensors, chips, and whatever else is left over.
John Koetsier retweeted
🔥🔥 Tesla Optimus Gen 3’s design has leaked through assets discovered in the Tesla Android app APK!
This quoted post is unavailable.
So @XPENG_Global says it will be shipping 1,000 humanoid robots per month by year-end
That would put it among global leaders for shipments.
The challenge: XPENG also says commercial deployments will start in 2027.
So I guess the question is: where are all the robots going?
🤖 Made with AI
John Koetsier retweeted
New humanoid entrant: Vesoma, out of stealth in Munich, Germany (plus a presence in Limassol, Cyprus).
Their bet: while most humanoid players train on human demonstrations, Vesoma is building a self-improving physical agent that learns from its own interaction, exploring, failing, and discovering what works for its own body. Competence grounded in physics, not examples, which they argue is what holds up when conditions change. They build their own body because owning it makes the learning fast.
The pedigree is the story. CEO Nikolai Ensslen co-founded Synapticon and developed the first safety architecture for humanoids. Chief AI Officer Martin Riedmiller is ex-Google DeepMind research director and a pioneer of data-efficient RL. Their designer shaped 3 generations of Tesla's Optimus and Apptronik's Apollo.
They claim a walking prototype in 6 months. No footage released so far. Funding undisclosed. Notable pledge: never for weapons, military, or surveillance.
John Koetsier retweeted
China Mobile just open-sourced a bridge between Physical AI models and humanoid robots.
China Mobile has released Open-RAIL, an open-source real-time execution layer for VLA and World Action Models .
It tackles a practical problem in humanoid robots:
The Physical AI model thinks in action chunks, while the robot body needs fast, continuous motion.
Open-RAIL sits in between:
VLA / WAM<->Open-RAIL <-> Humanoid Robot
It handles asynchronous inference, temporal alignment, and trajectory smoothing, turning low-frequency model outputs into smoother, high-frequency motion.
The project says it supports 20+ VLA/WAM models and 4 types of humanoid robots(Unitree G1/AgiBot G1/China Mobile Lingxi/NAVIAI-WA2), with tools for real-robot data collection and human intervention.
In real-robot testing, Open-RAIL achieved up to 2.09× faster task execution while reducing motion jitter.
Think of it as a real-time execution layer connecting robot AI models to humanoid robot bodies, making it easier to move different models across different humanoid robots.
What I think of when I see full massive 70-degree-of-freedom humanoid robots with legs and a head and complex hands being used to ... pick and place.
(This was at CES in 2019.)
John Koetsier retweeted
Here’s my outlook on humanoid robots. There are four stages of maturity, and each is gated by progress in the stage before it
First, you have to build great humanoid hardware
Second, that hardware needs to work with a whole-body, AI-first architecture
These first two chapters can’t be brute-forced with capital. They require deep engineering across hardware, controls, AI, and systems integration
The third chapter is scaling intelligence. The primary bottlenecks become data and compute. I believe humanoid robotics will ultimately require far more of both than LLMs
The final chapter is scaling manufacturing and integrating humanoids into the economy at massive scale
These last two chapters - intelligence and manufacturing - can be accelerated with capital. Doing them correctly will require tens, and eventually hundreds, of billions of dollars