RT by @huggingface: huggingface.co/collections/X…

TL;DR - Hugging Face amplified the release of "Xiaomi-Robotics-1," a XiaomiRobotics collection on the Hub centered on scaling Vision-Language-Action (VLA) models trained with 100K+ hours of real-world data. It signals a major industrial player publishing robot foundation-model artifacts openly rather than keeping them internal.

• Content is thin — essentially a collection link plus title — so specifics on architecture, model sizes, benchmarks, and license are not stated and cannot be inferred.
• The headline claim is a data-scale one: 100K+ hours of real-world (not simulated) robot interaction data used to train VLA models, positioning data scale as the primary lever.
• Distribution via a Hugging Face collection implies downloadable artifacts (models and/or datasets) grouped under a Xiaomi Robotics org, following the pattern set by other open robot foundation-model efforts.
• Placement as Industry rather than Opinion reflects that this is a corporate release announcement, despite arriving via a Twitter/X retweet.