LLMgram · AI News · 2026-08-09

Xiaomi open-sources Robotics-1 and its real-robot post-training pipeline

Xiaomi open-sources Robotics-1 and its real-robot post-training pipeline

Xiaomi has open-sourced Robotics-1, a robotics model the company says was pre-trained on more than 100,000 hours of UMI data and post-trained with over 10,000 hours of cross-embodiment data. The announcement says the release covers the pipeline from real-robot post-training through the model, which makes the implementation details more important than the headline training totals alone. The official X post does not specify the license, repository contents, supported robot embodiments, hardware requirements or evaluation results. Those details will determine whether outside teams can reproduce the workflow, adapt it to their platforms and compare the model fairly with other open robotics systems.

Sources

Xiaomi open-sources Robotics-1 and its real-robot post-training pipeline

Xiaomi open-sources Robotics-1 and its real-robot post-training pipeline

Xiaomi has open-sourced Robotics-1. The company says it used more than 100,000 hours of UMI data for pre-training and over 10,000 hours of cross-embodiment data for post-training. Xiaomi also says the release includes the pipeline from real-robot post-training through the model.

Key takeaway

The useful part of Xiaomi’s Robotics-1 release is the promised real-robot post-training pipeline alongside the disclosed training scale.

What happened

Xiaomi’s official technology account says Robotics-1 is now open source and reports more than 100,000 hours of UMI data in pre-training.

The same announcement reports over 10,000 hours of cross-embodiment post-training data and says the release covers the pipeline from real-robot post-training to the model.

Evidence

  • Xiaomi says Robotics-1 is now open source.

    Xiaomi Tech on X · official announcement

    Xiaomi-Robotics-1 is now open source!

  • Xiaomi reports more than 100,000 hours of UMI pre-training data and over 10,000 hours of cross-embodiment post-training data.

    Xiaomi Tech on X · official announcement

    Xiaomi-Robotics-1 was pre-trained on more than 100,000 hours of UMI data and post-trained with over 10,000 hours of cross-embodiment data.

  • Xiaomi says the open-source release covers the pipeline from real-robot post-training to the model.

    Xiaomi Tech on X · official announcement

    This open-source release covers the complete pipeline from real-robot post-training to model

Why it matters

Robotics teams need the training and adaptation workflow, not only a model artifact, to reproduce results across different hardware.

Limits and uncertainties

The official post does not specify the license, repository contents, supported embodiments, hardware requirements or evaluation results.

Practical implications

Teams can inspect whether Xiaomi published enough of the post-training workflow to adapt Robotics-1 to their own robot platforms.

What to watch

The repository, license, evaluation code and documentation for the claimed complete post-training pipeline.

Sources

LLMgram editorial selection and synthesis · @llmgram. LLMgram is not the original publisher of this information.
Continue on LLMgram: Open in AI Signal →
Original reporting: Xiaomi Tech