Morgan Stanley predicts that the world will have more than one billion humanoid robotics by 2050. Copyright AFP Jade Gao
Katie Forster with Luna Lin in Guangzhou
A pair of human-like robotic hands that swivel, built for artificial intelligence research in physical technology, mimic the motions of a VR operator spinning his hands like an illusionist.
Enactic, a Tokyo-based company, has developed humanoid robots that can wash dishes and do laundry for Japanese care homes, but they need more staff.
Welcome to the future as AI begins to permeate the material world through smart robots, autonomous cars and other machines.
Jensen Huang of the US chip giant Nvidia said last year that “the next wave of AI will be physical AI”.
Huang said that “AI that understands physics laws, AI that works among us”, and “understands how to perceive the world” is what he meant.
Morgan Stanley predicts that the world will have more than one billion humanoid robotics by 2050.
Videos of advanced androids, usually Chinese-made, dancing and pulling heavy objects are causing a lot of buzz.
Beyond the promise of scifi robot butlers the race has raised concerns over job loss, privacy, and how long it will take for these innovations to be useful.
Hiro Yamamoto, the CEO of Enactic and its OpenArm physical AI-training devices, is a 24-year-old man. Stanford University and Nvidia use these devices.
He plans to deploy new robots that are currently in development next summer, to “live alongside humans in environments where conditions are constantly changing and are chaotic”, such as care homes.
Yamamoto said, “So it must be safe”, with a soft exterior which won’t hurt anyone.
– ‘Any human role’ –
Last week, in the Chinese city of Guangzhou a female figure with an oval-shaped glowing visor for a head, dressed in white woven fabric as if she were a fencing athlete walked slowly across a platform to cheers and whispers.
The Chinese electric vehicle manufacturer XPeng has unveiled the latest humanoid robotic device.
The US has made headlines for its nimble robots. Boston Dynamics, for example, makes a dog-like robot.
Chinese rivals Unitree Robotics, EngineAI and others, including Unitree Robotics, are gaining ground thanks to government support and strong domestic suppliers.
He Xiaopeng, CEO of XPeng, told reporters that he had not given much thought as to how many robots he would sell in 10 years.
XPeng robots can walk, dance and even move objects autonomously. But how well they can handle objects is a more complex feat that has not been demonstrated.
He said that their dexterous fingertips and flexible skin will not replace workers in China’s factories anytime soon.
The cost of a robot hand that needs to be replaced frequently for heavy-duty tasks could pay for a Chinese worker’s salary for many years.
Brian Gu, co-president of XPeng, told AFP that AI humanoid robotics could perform “almost any role” in the future, from nanny, to chef, to gardener.
On-the-job Training
Text-based AI tools such as ChatGPT are trained using huge volumes of text, but physical AI models also need to grapple with vision and spatial relationships between objects.
Yamamoto said that for now, teaching AI robots how to pick up a cup by remotely operating them is “by far the most reliable method to collect data”.
Just 30 to 50 demonstrations of each task are needed to fine-tune “vision-language-action” AI models, he added.
Enactic has approached a dozen care facilities in Japan, proposing that its teleoperated robotics take over menial duties so qualified care workers can have more time to care for elderly residents.
Yamamoto said that this on-the-job training will train physical AI models to act autonomously in the future.
The US-Norwegian start-up 1X will be delivering its humanoid house helper NEO to American homes next year.
NEO costs $20,000, but its performance has been shaky. One video in US media shows the robot struggling to shut a dishwasher door even when teleoperated.
– Physical Limits
A Russian humanoid robotic, which was said to be its first, fell flat on its back as it made its stage debut earlier this week.
Sara Adela Abad Guaman is an assistant professor of robotics at University College London. She said that there is a “big” gap between the AI systems of robots and their physical capabilities, which are lagging behind.
“Nature shows us that to adapt to an environment, you must have the right body,” Abad said to AFP, citing the example of a goat on ice.
Despite the fears of a stock bubble, large deals are still being made, despite the booming investment in AI.
SoftBank, a Japanese company, recently called physical AI the “next frontier”, and announced that it had purchased industrial robot maker ABB Robotics from Japan for $5.4 billion.
Abad isn’t too concerned about the future of human labor, despite the fact that automation raises questions.
She said that “our senses of touch are incomparable” at the end of the week.
