Xiaomi's Robotics Model Xiaomi-Robotics-1 Breaks the Data Barrier
Revolutionizing robotic capabilities with large-scale data training.
1The Data Barrier in Robotics
The lack of large-scale, high-quality data in robotics has limited the scalability of policy models. Xiaomi-Robotics-1 combines large-scale pre-training with real-robot data to overcome this barrier. This approach aims to expand the boundaries of robotic capabilities.
2Data Collection and Processing
Xiaomi-Robotics-1 uses 100,000 hours of pre-training data and 7,200 hours of real-robot data. Data is collected from household, commercial, and industrial environments. An auto-labeling pipeline enriches the data with language-based scene transitions, enabling training across a wide range of tasks.
3Training Process
The model undergoes a two-stage training process: pre-training and post-training. Pre-training learns general representations from a broad data range, while post-training ensures real-robot alignment and instruction-following. This process ensures success in real-world tasks.
4Scalability and Success
Xiaomi-Robotics-1 demonstrates that success rates improve steadily as pre-training data and model size increase. This proves scalability is achievable in robotics. The model also achieves top results in simulation benchmarks.
5Application Areas
The model can learn new tasks with just a few hours of data. It achieves a 75% success rate in complex tasks like phone packing and laundry loading. This boosts efficiency and reduces costs in robotic applications.
6A New Era in Robotics
Xiaomi-Robotics-1 pushes the limits of robotic capabilities with big data and scalable models. This breakthrough has the potential to revolutionize household and industrial robotics, paving the way for smarter and more capable robots in the future.