Paxton says Mind-1's fast videos look like UMI, not teleop
One researcher's read on a humanoid video that MindOn has not explained.
MindOn posted Mind-1, which it pitches as physical AI at human speed, built for real work. The company's framing is that finishing a task is the easy part and doing it at a pace that works in the real world is the next step.
Chris Paxton watched and wanted the engineering. He said the videos are consistently mind blowing, with very fast, smooth, humanlike motion, and that he really wants to know more about how the stack works.
The guess that matters
Paxton's read is that this is UMI-style work, not teleop data. UMI-style means the policy learns from demonstrations a human collects with a handheld rig, rather than from teleop (a person driving the robot remotely while the data is recorded). Teleop data tends to be slower and more deliberate because a human is fighting latency and an unfamiliar control scheme. Human-collected handheld demos run at human speed, which is one reason the resulting motion can look quicker and less robotic.
That is why the guess is interesting rather than trivia. Speed is the thing almost every humanoid demo is bad at, and if Mind-1 got it from how the data was gathered rather than from a faster controller, that is a repeatable recipe other labs can copy.
Shown versus claimed
Nothing here is confirmed. Paxton is one researcher reading a video, and he says he is fairly sure, not certain. MindOn has not said how the robot is controlled, what model is running, whether the clips are autonomous, or whether the footage plays at real time. The company's own post is a tagline and a video, not a technical disclosure.
So the honest summary is that a well known robotics researcher looked at a humanoid moving fast and smoothly and thought the training data explained it. Whether he is right is up to MindOn, and so far MindOn has not said.
Fairly sure this is UMI-style work, not teleop data.
very very fast and smooth, humanlike motion