Xiaomi has released an AI model for robots, Xiaomi-Robotics-1, that follows the same scaling pattern as large language models. Its performance improves as it trains on more data. To build that dataset, Xiaomi largely avoided using physical robots. Robot AI faces a data problem that language models don't: LLMs can train on huge parts of the public internet, while useful data on robot movement is scarce. Robots usually have to learn from scratch how to grip, lift, and put away objects. The standard method has people remotely guide a physical robot through every movement. That process is slow and expensive, and it often produces repetitive data from the same tasks in the same settings. That's the gap Xiaomi-Robotics-1 is trying to close.
To get around the data bottleneck, Xiaomi mostly ditched real robots during data collection. Instead, the team used portable handheld grippers with attached cameras that a person simply picks up and operates by hand. This setup lets you record manipulation tasks in kitchens, offices, stores, factory floors, and outdoor spaces without a robot even being present. The result was over 100,000 hours of motion recordings. The pretraining data comes from about 100,000 hours of UMI recordings across more than 1,700 different environments. Labeling it all by hand wasn't practical, so Xiaomi used another AI model to describe each motion segment in text. The team says it labeled the full dataset in about two weeks.
Xiaomi then transferred that training to physical robots, including wheeled models and dual-arm systems. The model still had to account for the differences between a handheld gripper and a robot arm. For post-training, Xiaomi combines its own recordings from real apartments with open-source robot datasets and annotated UMI data. More training data matters more than model size. Tests showed that a larger model improved performance, but more training data produced much bigger gains than more compute. The researchers say progress in robot AI will depend mainly on collecting larger and more varied datasets.
Source: thedecoder
