Xiaomi-Robotics-1: Scaling Robot Policy Models with Large-Scale Data
Xiaomi-Robotics-1 leverages large-scale data to scale robot policy models, showing improved performance in real-world tasks.
Xiaomi-Robotics-1 combines large-scale embodiment-free (UMI) pre-training with real robot data to enhance robotic capabilities. The model utilizes 100,000 hours of diverse data for pre-training, aiming to break the limitations of traditional robotics. Results indicate that the scaling behavior observed during training translates effectively to real-world robot performance.