Xiaomi-Robotics-1

The conversation settled on a clear split. Optimists saw the same pattern as early robot vacuums, dishwashers, or even early cars. Slow and imperfect still wins if it works unattended, overnight, and often enough. A badly folded shirt is acceptable if the chore disappears. Skeptics pushed back that this is still demo-land. The reported task success rates are far from product reliability, controlled environments matter a lot, and robotics has a long history of eye-catching clips that never become robust consumer systems. Several people pointed out the ugly truth of robotics economics too. Getting from an 80 percent solution to the last 20 percent can take forever, and for household work a human cleaner is still cheaper than any plausible near-term humanoid. A few technical points sharpened the picture. People familiar with the field noted that these modern full VLA systems do not build an explicit world model in the traditional robotics sense. They learn internal representations implicitly through end-to-end training, which is part of why they look more like scaled AI systems than hand-built state machines. Others highlighted that the small model size by language-model standards is encouraging, and that the "embodiment-free" setup appears to mean some portability across robot arms that share a standard gripper-camera arrangement, even if real deployment would still need fine-tuning. The wider mood was upbeat about the research and the open-weight direction, but much cooler on timelines. The strongest practical read was simple: household dexterity is getting real enough to matter, but nobody should confuse a good laundry video with a buyable home butler.