Xiaomi Open-Sources Robotics-1 Foundation Model Trained on 100K+ Hours of Data
burny_tech · x · 2026-08-06
Xiaomi has officially open-sourced Xiaomi-Robotics-1, a foundation model for robots that adopts an LLM-style training paradigm to overcome the scarcity of high-quality data in robotics.
- Data Scale: The pre-training phase utilized over 100,000 hours of embodiment-free (UMI) trajectory data across 1,700+ scenarios. The post-training phase incorporated 10,000 hours of cross-embodiment data, including 7,200+ hours of real-robot data collected in real homes.
- Methodology: Developed a scalable auto-labeling pipeline that segments trajectories and annotates them with language descriptions of scene state transitions.
- Release: The open-source release covers the complete pipeline from real-robot post-training to deployment, along with evaluation code for benchmarks.
Related event: Xiaomi Releases and Open-Sources Robotics Foundation Model XR-1(6 posts)→
More from Embodied
- Ropedia raises tens of millions in funding to build real-world experience infrastructure for Physical AI — 机器之心 · 2026-08-06
- SJTU Team Proposes GSR to Tackle Instruction Generalization in VLA Models — 机器之心 · 2026-08-06
- Overcoming Robot Diffusion Policy Latency: Sub-ms Inference via Action History Initialization — chris_j_paxton · 2026-08-06
- MARS Policy: Zero-Latency Multimodal Robot Actions via Adaptive Noise Injection — chris_j_paxton · 2026-08-06
- Echo Show Jailbroken to Android Runs Smoothly, User Suspects Planned Obsolescence — brianrkelly · 2026-08-06
- Meituan drones surpass 1 million deliveries, opens low-altitude logistics solutions to partners — 美团技术团队 · 2026-08-06