July 20, 2026
Bots, chores, and comment chaos
Xiaomi-Robotics-1
Xiaomi’s new robot brain drops, and the comments instantly ask: can it fold laundry yet
TLDR: Xiaomi says its new robot model learned from 100,000 hours of real-world action, a big step toward robots that can handle everyday tasks. The community reaction was peak internet: part impressed, part skeptical, and heavily focused on the only metric that matters—whether it can finally do the laundry.
Xiaomi just unveiled Xiaomi-Robotics-1, a new robot “brain” trained on a jaw-dropping 100,000 hours of real-world footage showing hands, grippers, and objects being moved around in homes, shops, factories, and outdoors. Xiaomi says this giant pile of experience helps the system learn general skills first, then adapt to real robots with another 7,200 hours of actual robot practice in real homes. In plain English: it watched an absurd amount of stuff being picked up, sorted, and put away, and now Xiaomi wants us to believe the robot future is finally getting practical.
But the real show was in the comments, where readers immediately swerved from corporate demo hype into classic internet cross-examination. One camp was impressed and curious: can this thing work on robot bodies it hasn’t seen before? Another wanted to compare it to the bigger self-driving names, basically asking whether Xiaomi is playing in the same league as Tesla and Waymo. And then came the delightfully petty, deeply human reaction that stole the thread: if robots are so smart, why are we still not getting automated ironing and folding? One commenter basically shrugged at suitcase packing and demanded the chores people actually hate.
There was also a weirdly charming side quest: someone was almost more impressed that Xiaomi’s page worked without JavaScript, which is the kind of nerdy compliment that lands like applause in tech circles. Add in a comment noticing the robots seemed to cooperate, and the mood became a mix of awe, skepticism, and “cool story, now do my laundry.”
Key Points
- •Xiaomi says Xiaomi-Robotics-1 is a ready-to-use robot foundation model trained on over 100,000 hours of real-world manipulation trajectories.
- •The pre-training dataset consists of 100,000 hours of embodiment-free UMI trajectories across more than 1,700 scenarios in household, commercial, industrial, and outdoor environments.
- •Xiaomi built an automatic annotation pipeline using a vision-language model to label fixed-length trajectory segments with language descriptions of scene state transitions.
- •For post-training, Xiaomi used cross-embodiment datasets including over 7,200 hours of in-house real-robot data collected in real homes, plus filtered open-source robot data and high-quality UMI data.
- •Xiaomi reports that during pre-training, validation action error decreases as both data scale and model size increase, and post-training is used to align the model with real robot embodiments and instruction following.