More Data Helps Robots Learn Faster

Xiaomi trained their new robot system using over 100,000 hours of movement data. Instead of using robots to gather this, they had people use handheld grippers with cameras. This approach taught the machine how to move more effectively.

The team discovered that adding extra data helped the robot learn more than simply making the underlying model bigger. The performance kept improving as they added more examples to the system.

While the robot is getting smarter, it still makes mistakes often. Success rates are currently low but this research proves that data is the key to better movement.