DistantNews
🇰🇷 South Korea /Technology

China's 'Robot School' trains humanoid robots with massive data for smarter, more capable machines

From Hankyoreh · (7m ago) Korean Positive tone

Translated from Korean, summarized and contextualized by DistantNews.

TLDR

  • China is rapidly advancing its humanoid robot industry by establishing large-scale data training centers, such as the Beijing Humanoid Robot Data Training Center.
  • These centers collect vast amounts of data from robots performing various tasks, which is then used to train AI and improve robot intelligence and dexterity.
  • While progress is rapid, the widespread adoption of domestic robots faces hurdles such as high costs and current limitations in AI capabilities, though rapid evolution is expected.

China's ambitious push into the humanoid robot sector is gaining significant momentum, exemplified by the sophisticated operations at the Beijing Humanoid Robot Data Training Center. This facility, a 'robot school' in essence, is at the forefront of generating the massive datasets crucial for advancing artificial intelligence and enhancing robot capabilities. As reported by Hankyoreh, the center's focus on collecting real-world interaction data—from folding towels to preparing toast—is key to making robots more adaptable and human-like.

Humans can do this after a few tries, but robots need tens of thousands of practice sessions. If the towel's material changes, they have to learn all over again.

— Liu ZhongxunManager at Realman Robotics, explaining the intensive training required for robots.

The scale of this initiative is immense. With the capacity to produce 6 million data points annually, the center utilizes 108 robots engaged in daily training. This data, encompassing up to 71 dimensions per action (including visual, tactile, and temporal information), is meticulously gathered and processed. The underlying philosophy is that real-world data, far more valuable than simulated environments, is the essential ingredient for developing robots capable of navigating complex domestic settings.

The most valuable data is that produced in real environments.

— Liu ZhongxunManager at Realman Robotics, emphasizing the importance of real-world data collection.

While China's progress in robotics is undeniable, with companies like Realman Robotics leading the charge in hardware and data collection, the path to mass consumer adoption is still being paved. The current high costs of domestic robots and the nascent stage of their AI intelligence present challenges. However, the rapid improvements observed—such as a robot's ability to fold a towel smoothly now compared to a year ago—suggest that the evolution of robotic capabilities and their integration into daily life will likely accelerate significantly.

One robot action data point contains information in 71 dimensions, including vision, touch, trajectory, and time.

— Liu ZhongxunDetailing the complexity and richness of the data collected for robot training.
DistantNews Editorial

Originally published by Hankyoreh in Korean. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.

More Stories

From Our Blog