code_weights_and_data_release_state · verified_hold
审核码：verified_hold; 官方关系：verified_official_placeholder_after_arxiv_v2; 最近核验：2026-07-23代码：Fixed commit 09a2eb41b8ccc4a4236a66f9119d6854321ca814 contains only README.md and the paper PDF.
权重：The official XiaomiRobotics Hugging Face organization has no Xiaomi-Robotics-1 model repository.
数据：The >100K-hour UMI and about 10K-hour post-training corpora have no public files, manifest, schema, samples or license.
许可证：not_disclosed；The placeholder repository has no LICENSE; the paper license does not grant code, model or dataset reuse.
依赖：未披露
硬件/传感器：Undisclosed in-house mobile and dual-arm embodiments、Training GPU count, inference latency and minimum VRAM not disclosed
实机证据：Four downstream tasks use ten trials each and report 75% average success in the 36-hour low-data setting.；Four out-of-box tasks report aggregate scaling results, but the public system has no deployable artifact.
独立复现：No independent result reproduction exists; public GitHub issues request the missing UMI corpus and checkpoints rather than reporting a reproduction.
限制/反面证据：The full 100K-hour corpus is not the same as the approximately 20K-hour subset used in scaling experiments.；No public training recipe, checkpoint, supported-hardware matrix, inference latency or license exists.；All physical evidence is first-party.
当前建议：高潜待持续验证；Treat the paper as a high-priority scaling result and watch target, not as an adoptable model or dataset until public files, licenses and a bounded deployment path exist.
重点标签：通用VLA规模化预训练主流迭代、artifact:placeholder、data:100K-claimed-20K-scaling-subset、weights:not-released、community:unverified
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4