A pair of human-like robotic arms swivel and mirror the movements of an operator in a VR headset, offering a glimpse into the growing world of physical artificial intelligence. These robotic arms, built for advanced research by Tokyo-based company Enactic, are part of a larger vision to create humanoid robots capable of handling everyday tasks independently. Enactic believes that with enough training, such machines will soon wash dishes, do laundry and support short-staffed care homes across Japan.
Tech leaders and researchers are embracing this shift as AI begins to move beyond screens and into physical environments. Jensen Huang, head of Nvidia, captured the moment last year when he said the next wave of progress is “physical AI”, referring to systems that can understand physics, operate safely around humans and interpret the world clearly. Investment is rising rapidly, and Morgan Stanley predicts that the world could have more than a billion humanoid robots by 2050.
Alongside serious scientific work, the hype continues to grow as videos circulate showing advanced androids, many from China, dancing to Taylor Swift or dragging heavy objects with surprising ease. While these clips spark excitement, they also raise concerns about job displacement, safety and how long it will take before such robots become genuinely useful in daily life.
How are companies preparing robots to work alongside people?
Hiro Yamamoto, the 24-year-old CEO of Enactic, is leading efforts to make robots that can live and function in unpredictable spaces such as care homes. His company’s OpenArm training devices are already used by Nvidia and top universities including Stanford. Yamamoto said the new humanoid robots planned for release next summer must be safe enough to operate in chaotic, ever-changing environments, noting the need for a soft exterior to protect people.
In China, the push for physical AI is equally aggressive. In Guangzhou, XPeng unveiled its newest humanoid robot to cheers as it walked across a stage wearing white woven fabric and a glowing visor. The company joins others like Unitree Robotics and EngineAI in a fast-growing field supported by strong domestic supply chains and government backing. XPeng CEO He Xiaopeng said he believes the company could one day sell more robots than cars. While their machines can walk and dance autonomously, handling objects remains a challenge, especially since maintaining robotic hands can cost more than employing human workers for years.
XPeng’s co-president Brian Gu sees potential far beyond factories. He said humanoid robots could eventually perform “almost any human role”, from caregiving to cooking or gardening, once they gather enough training data and improve their physical capabilities.
What challenges stand between physical AI and large-scale adoption?
Developing physical AI requires teaching machines how to process vision, movement and spatial awareness. Yamamoto explained that remotely operating robots remains the most effective way to gather training data for actions such as picking up a cup. Around 30 to 50 demonstrations per task are enough to refine complex “vision-language-action” models that guide the robots.
Enactic is already in talks with dozens of Japanese care homes to allow teleoperated robots to manage simple chores. This real-world work would help the robots learn autonomously over time. A similar strategy is used by US-Norwegian startup 1X, which plans to deliver its home helper robot NEO to American households next year. Despite its $20,000 price tag, early demonstrations show that NEO still struggles with basic household tasks, even under human control.
The limits of physical AI remain clear. A Russian humanoid robot recently fell flat on its face during a public debut, and experts note a big gap between advanced AI software and the mechanical bodies that must execute those decisions. University College London robotics professor Sara Adela Abad Guaman said, “Nature has shown us that to adapt to the environment, you need to have the right body,” pointing to the example of a mountain goat slipping on ice.
Still, major deals reflect growing confidence. Japan’s SoftBank has declared physical AI its “next frontier” after announcing a $5.4 billion purchase of ABB Robotics. Questions remain about automation and the future of human labour, but Abad believes people still hold a major advantage. As she put it, “our sense of touch is incomparable”.
By Lucky Anyanje