/Tech15h ago

Robotics Veteran Warns Physical AI Hype Masks Hard Technical Challenges

2127187226.9K
ControlWiz
kache#487
Original postkache#487
ControlWiz@Control_wiz

I’ve been in robotics for about 7 years. For most of it, especially in controls, it was a pretty niche field. Now it seems like “physical AI” is the thing to do.

I’m glad robotics is getting all this attention, and we need more people to tackle these problems. But I also think it's overhyped and misleading. A lot of the spotlight is on curated demos or robots dancing/doing flips. They’re cool, but they make the field seem much further along than it really is.

The hard problems, robust manipulation, sim2real, long horizon autonomy, reliability and more, don’t make for a cool 30 sec demo.

yushjinhun@yushijinhun

I'm not an opportunist. I chose robotics because it's robotics, not because it's the hot field. My passion will not bend to the hype.

8:05 AM · Jun 10, 2026 · 21.9K Views
Sentiment

Many users voiced skepticism about physical AI and humanoid robot hype, criticizing transformers for poor reliability, control theory's failure on edge cases, and lazy designs with inadequate hand and foot dexterity.

Pos
0.0%
Neg
100.0%
4 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS6.9KBOOKMARKS6
leepy@leepy_cat

@Control_wiz egoscale solved sim2real btw

12hViews 6.9KLikes 2Bookmarks 6
LIKES73REPLIES6
kache@yacineMTB

I'm having a lot of trouble with mujoco . Not because it's bad, it's actually quite good. But because it's hard. This is hard lol

3hViews 5KLikes 73Bookmarks 3
RETWEETS1
Andrew — e/acc@astralmatrix

@Control_wiz physical AI, or AI-driven robotics became the thing once LLMs became capable enough. it's a very different problem to solve than fabricating robotics, especially humanoid robotics. the distinction is important bc the two tracks are running in parallel at the moment

11hViews 197
kache@yacineMTB

@Control_wiz we need more easily buildable robots and baselines, remote control hobby tier 3d printed crap

14hViews 1.6KLikes 47Bookmarks 2
kache@yacineMTB

Programming in general is hard

3hViews 348Likes 9
Giskard@giskard2

@Control_wiz @yacineMTB We need a new approach - transformers are versatile but they suck at reliability without harness engineering- feel like a dead end given the data constraints.

It’s fundamentally a problem of robust long horizon agency.

13hViews 196
Ax🕸️@wireheaded

@Control_wiz Isn't figure ai's helix engine trying to solve these exact long horizon tasks. What's your opinion on them and their approach

10hViews 41
patrick@ConsumerRick

@Control_wiz control theory seems to solve a lot of problems with repetitive processes, but basically cannot handle edge cases to save its life

9hViews 4.2K
Chandra Shekhar 🛡️@cspaliwa1

@Control_wiz Hey I am amature is this

Could you list all "important" problems in simulation of robots?

I rn work on infinite edge cases

13hViews 255
wildiris@wildiris19

I share your skepticism. One must first leave the engineering lab, go out to the field, and witnesses skilled human labor in practice; ..only then does it hit how far humanoid robotics still have to go. • Comparing videos of human field workers harvesting celery to videos of humanoid robots we've seen lately, ..makes the robots seem like they're perpetually moving in slow motion. • The video linked below is from Salinas, CA. Just a short drive south from where I live. • Because I see this all the time, I've never been able to get too excited about humanoid robots. They are still light years away from competing with skilled human labor. • The truth is, that farm field labor will be some of the last people, if ever, to be replaced by robots.

Increíble cosecha de apio por contrato en USA / Celery Harvest in Americ... https://youtu.be/-Z1RwPKMuOk?si=StN6dr0aslsyKZ-u via @YouTube

9hViews 154
Matt DeMonte@mattdemonte

@Control_wiz There is a finding in neuroscience that in fact most of our brain is just being used to move our bodies around in space and use our hands etc. That is a bit of cold water when you think about what parts of our brain have been decently simulated (mostly language.)

13hViews 147
??????????????🐍@NabZO560

@Control_wiz can software help robotics ? whats missing there ? ( aside from AI )

14hViews 140
@nickrackley@nick_rackley

@Control_wiz Do you actually believe in sim2real?

I was really focused on doing continuous real life reinforcement learning on robots, it seemed like just as likely to work out if you scaled enough ss sim2real. I didn’t have the resources to burn up as many bots as I would need though.

12hViews 92
Maclaine@macmaniac77

@Control_wiz Though it shouldn’t be necessary I would love to see a self correcting palletizer model/control capability.

4hViews 9
Tommy Ly@lymytom20

@Control_wiz american dynamism = defense tech physical ai = robotics world model = Pearlian causal model high agency = motivated taste = vision

the media tends to invent new term for old vocab

3hViews 5
Cooper@Jake32651324

@giskard2 @Control_wiz @yacineMTB Sentence transformers had the same problems till they got LLMs

10hViews 5
NCD@Jokerbernrin

@yacineMTB The learning curve is real

3hViews 4
Mallchad 🏴󠁧󠁢󠁥󠁮󠁧󠁿@Mallchad

@wireheaded @Control_wiz Well for a start they decided to be lazy on the hand and feet design... Which appear to have no splay and minimal dexterity, poor grip surfaces, little adjustability...

So that does not bode well for a "humanoid" robot. Why even bother with humanoid if you don't want to do hands

8hViews 3