The LLM can do only so much. Robots urgently need a new model which understands physical reality the same way as a LLM would understand written words. Engineers need more information to create this model. As I walk around BAAI, there are scores of people behind desks. The robots are teleoperating different grippers and arms to train algorithms how to perform simple manipulation tasks, such as picking up items off shelves, pouring liquids into cups from a large jug, or sweeping beans from a table. In this video, a man with a VR headset is seen making tea while a camera captures his movements. It is believed that robots can learn to perform all kinds of tasks without any training if they are given enough data.
Problem is that nobody knows which data will be most helpful to robots. Nor do they know how to best collect or store it. In order for humansoids to be ubiquitous, hardware must better mimic a human’s hand. A robot can do a better job of flipping coins than a human.
Tony Zhao is the cofounder and CEO at Sunday Robotics. He’s from California. Tony Zhao tells me that his company doesn’t have a good chance of competing with Chinese startups, as they can employ more people, like BAAI teleoperators. They are able to rapidly deploy new hardware and quickly train their robot models. “The iteration speed, the US is losing there,” He says. “And honestly I don’t know how we can win.”
Zhao, in an effort to remain competitive, recently hired a Chinese executive who had deep contacts and extensive experience with China’s massive and complex supply chains. “The only way we can beat Chinese companies is to build a China team,” He says.
Demis Hassabis, of Google DeepMind, and Lachy Groom, of Physical Intelligence (both of whom are racing to capture the roboChatGPT moment) have both told me that their vision of robotics is similar to the development of smartphones. China would make the hardware while the US would produce the brains. Huawei now makes both.
Jonathan Hurst suggests that US Government should get involved. Agility makes humanoids. As a means to support domestic robotics companies, he envisions, among others, heavy investments in advanced manufacturing at home, including tax incentives for firms using robots in warehouses and factories. A similar strategy would be akin to the Chinese government investing patiently in its industries. “We have to be very smart about automation,” He says. “It is the only way.”
The wheeled robots which deliver the items in many big city hotels to guest rooms are not present in my hotel, located in Zhongguancun. Stephen was a polite and always helpful human. Stephen was able to clean my shirt in just hours. When I returned home after my vacation, I thought about how many people had worked so quickly to wash, press, package, and transport the garment. Robots are still not winning in China.
What Say You?
Comment below to let us know your thoughts on this article. Alternatively, please send an email to the Editor at mail@wired.com.
