Xiaomi-Robotics-1 shows that more data beats bigger models when training robots to move
Quick Answer
Xiaomi-Robotics-1 demonstrates that more training data significantly enhances robot performance, achieving a 75% success rate in unfamiliar tasks with just 10 hours of training per task, outperforming competitors.
Quick Take
The model uses over 100,000 hours of motion recordings from handheld grippers, emphasizing data over model size for advancements in robotic AI.
Key Points
- Xiaomi-Robotics-1 achieved a 75% success rate in unfamiliar environments with limited training data.
- The model was trained on over 100,000 hours of motion recordings from handheld grippers.
- Xiaomi's approach emphasizes data collection over model size for robotic AI advancements.
- The model outperformed a competitor, achieving 75% success compared to 40% in similar tasks.
- Xiaomi plans to release the model and code on GitHub and Hugging Face.
Source Excerpt
Xiaomi trained Xiaomi-Robotics-1 on more than 100,000 hours of motion data collected by people using camera-equipped handheld grippers rather than robots. Adding data improved performance far more than increasing model size. The gains haven't plateaued, though absolute success rates remain low.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from The Decoder
See more →
An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run
Epoch AI's MirrorCode benchmark reveals Claude Opus 4.7 as the leader with a 56% solve rate, reconstructing a 16,000-line toolkit in 14 hours. Despite this, all models tested struggle with the most complex tasks, highlighting limitations in current AI capabilities. The single task consumed $2,600 over 19 days, raising questions about cost-effectiveness in AI development.
