Paxton teleoperates Feather's $33,000 semi-humanoid
He drove it by hand, which tells you about the hardware and nothing about autonomy.
Chris Paxton got hands on the controls of a Feather Robotics semi-humanoid and posted what he found. His numbers, straight from him: $33,000, 15 lb payload per arm, human sized with what he calls fantastic reach. His verdict was that it seems like a great and affordable industrial semi humanoid.
It always gives a clearer (and usually more positive) view of what a robot can do
All around seems like a great and affordable industrial semi humanoid.
The important word is teleoperating (a human driving the robot remotely). Paxton was the one moving the arms. Nothing in his post says the machine did any of it on its own, so this is a read on the hardware, not on a policy or a model.
Why a hands-on read matters
Paxton makes the case himself for why driving a robot beats watching a video of one. In his words, it "always gives a clearer (and usually more positive) view of what a robot can do." A demo clip hides latency, slop in the joints, how far the arms actually reach and how the thing behaves when you ask it for something awkward. Someone at the controls feels all of that in a minute.
That cuts both ways here. The $33,000 price and the 15 lb per arm payload are figures Paxton reports after a session with the team, not results from an independent test. Nobody has published a throughput number, a duty cycle, or a task success rate for this robot.
Shown versus claimed
What was shown is a human sized two armed machine that a visiting researcher found pleasant and capable to drive, at a price well under what industrial humanoids usually carry. What was not shown is the robot doing a job by itself. Those are separate problems, and the second one is the one everybody in the field is still fighting over.
Paxton has been one of the sharper public skeptics this month, waving off a paper that claimed embodied models could be tricked into stabbing someone and poking holes in a fast interactive world model. When he says a robot is good, it is worth a look. He also said he was really glad the team let him try it.
Sound like you were at the demo
One email a day. Under three minutes. The demos, the deployments and the part the press release left out.