Pinto: Astra, Fable, Muse have academics in despair
Lerrel Pinto says academics are despairing as Astra, Fable and Muse zero-shot robotics benchmarks, Ken Goldberg calls the shift thrilling, and Chen Tessler still cannot reproduce the Astra results. Plus UBTECH's humanoid factory and an Optimus V3 hype check.
The big stuff
Pinto says academics despair as Astra, Fable, Muse zero-shot robotics benchmarks
Lerrel Pinto put words to the mood: academics are in despair because Astra, Fable, and Muse are zero-shotting (solving with no task-specific training) benchmarks in robotics and world models, and he says this is what step jumps in progress look like. Ken Goldberg replied that he hears it but finds the paradigm shift thrilling, and Chris Paxton said the soul searching is not limited to academia. Paxton had also posted separately that everyone is still Astra-posting while he keeps finding problems it cannot solve, and Chen Tessler compared the Astra posts to pizzas he made: they look amazing but he has no idea how to reproduce them.
Uneasy pill to swallow, but this is what step jumps in progress looks like.
I hear you. But also thrilling to witness a paradigm shift in robotics.
They look amazing, but I have no idea how to reproduce them.
Deployments and money
UBTECH opens 10,000-unit humanoid factory, robots kit and fetch for humans
@humanoidsdaily reports UBTECH has commissioned a 10,000-unit-scale factory in Liuzhou, Guangxi, to build its Walker S humanoids. The 14,000 m² plant is engineered for a 10-minute takt time (one unit coming off the line every 10 minutes) and runs a flexible, non-conveyor line where mobile manipulators and AGVs (automated guided vehicles, wheeled robots that carry parts) handle kitting (gathering the parts for a job) and logistics. Chris Paxton's take: the robots are doing the easy stuff and bringing parts to human workers, which is how you know we are not heading for a hard takeoff yet.
Engineered for a 10-minute takt time, the 14,000 m² facility features a flexible, non-conveyor assembly line where mobile manipulators and AGVs handle kitting and logistics to build the Walker S series.
To be honest the reason you know we aren't heading for hard takeoff yet, is that the robots are kitting and doing easy stuff and bringing things to human workers
Discourse and drama
Humanoid Scott says Optimus V3 secrecy claim is four-month-old news
The Musk lines going around, that competitors do frame-by-frame analysis and that there is value in not showing new technology until it is closer to production, come from Tesla's Q1 2026 earnings call on April 22, over four months ago, Humanoid Scott points out. He also resurfaced his earlier argument that no one, including Tesla, has a humanoid prototype worth scaling in the tens of thousands, and that V3 will not be shown this year because the bar for humanoid demos is beyond what it can do, not over piracy concerns. Rob Forth added that Musk's old narrative about prototypes being the easy part no longer holds for humanoids, and Scott agreed.
This is not news to anyone following the space
It’s not a scaling problem, is a prototype problem
In the past Musk's narrative has always been that prototypes were the easy part. With humanoid robots even that's no longer true.
Quick hits
Paxton calls Asimov's zero-shot sim-to-real an important achievement
Chris Paxton pointed to zero-shot sim-to-real (a policy trained only in simulation that runs on the real robot with no tuning) for a robot called Asimov and called it an important achievement. His point: sim-to-real only looked easy on the Unitree G1 because the whole international community was working on that one robot.
Simon Kalouche posts a Unitree G1 teardown video
Simon Kalouche posted a teardown of the Unitree G1 humanoid, with a link to the full video. If you want to see what is inside the robot everyone trains on, this is the one to click.
Chris Paxton says self-driving trucks are still chugging along despite the mood
Chris Paxton wrote a blog post arguing self-driving trucks are still chugging along, which he admits is not the vibe on X right now. Anthony Chen replied that the driver was doing a job nobody wrote down and removing the driver exposed it, and Paxton agreed: across industries, human workers do far more tasks than advertised, even in tightly bounded roles.