But language models can only do so much, and what robots urgently need is a fresh kind of model that understands the physical world in the same way LLM understands the written word. To build such a model, engineers need much more data. Walking around BAAI, I see dozens of employees behind desks. They teleoperate various robotic arms and grippers to teach algorithms basic manipulation tasks such as sweeping beans on a table, pouring liquids from a jug into different cups, and selecting items from shelves. A adolescent man wearing a virtual reality headset appears to be making a cup of tea while the camera records his every move. The idea is that given enough training data, robots will intuitively sense how to do things without special training.
The problem is that no one really knows what data is most useful for robots, let alone how much they need and how best to collect it. For humanoids to become ubiquitous, humans must invent equipment that better mimics the human hand. For a robot, doing a backflip is much easier than flipping a coin.
Still, Tony Zhao, co-founder and CEO of Sunday Robotics, a California startup, worries that companies like his have little chance compared to Chinese companies that can hire more workers like teleoperators BAAI to train robot models and quickly deploy fresh equipment. “The U.S. is losing speed of iteration,” he says. “And honestly, I don’t know how we can win.”
To keep pace, Zhao recently hired an executive from a Chinese robotics company with deep connections and experience in China’s extensive and sophisticated supply chain. “The only way we can beat Chinese companies is to build a Chinese team,” he says.
Some U.S. CEOs, including Demis Hassabis of Google DeepMind and Lachy Groom of Physical Intelligence, both chasing the robo-ChatGPT moment, told me they envision the development of robotics roughly mirroring that of smartphones, with China producing the hardware and the United States producing the brains. (Except Huawei now produces both.)
The answer may be that the U.S. government should get involved, suggests Jonathan Hurst, co-founder and chief robotics officer at Agility, a humanoid manufacturing company. Among other things, he envisions huge investments in advanced domestic manufacturing, such as tax incentives for companies that exploit robots in their warehouses and factories, as a way to support domestic robotics companies. Such a strategy could begin to mimic the Chinese government’s patient capital investment in its industries. “We have to be very smart about automation,” he says. “It’s the only way.”
At my hotel in Beijing, in the contemporary center of Zhongguancun, there was no rolling robot that routinely delivers items to guest rooms in some hotels in massive cities. Instead, an unfailingly polite man named Stephen had mine. When I needed to neat a shirt, Stephen did the job in just a few hours. As I returned home at the end of the trip, I wondered how many hands had washed, ironed, packed, and transported the clothes back at such speed. Even in China, robots have not yet won.
What do you say?
Let us know what you think about this article in the comments below. Alternatively, you can send a letter to the editor at mail@wired.com.