In my last Last Word, I cast shade on AI due to the fact that it is unable to make a cup of tea. However, the burgeoning field of ‘Physical AI’ means that I was already behind the times. It turns out that making tea is probably the least of the tasks that AI will be able to accomplish in the real world.
Physical AI is the reach of artificial intelligence into the real world, through interaction with real objects. This will be accomplished through any number of means, but primarily through robots, a term first used in 1920 by Czech writer Karel Čapek in his play R.U.R. (Rossum’s Universal Robots). Now, more than 100 years later, we are seeing an acceleration in development of practical, robust, mobile and economical robots - in a wide variety of forms.
In my increasingly distant youth, I used to watch Dr Who, a British sci-fi series. There were two main types of villains: the Daleks (robot-like beings that trundled around on wheels and which would not have coped well if confronted with stairs) and the Cybermen (humanoid cyborgs for which stairs held no fear). These two enemies of Dr Who embody the robot forms currently being developed: those that are only required to cope with flat floors, for which locomotion is provided by wheels, and with a more-or-less human-looking tool platform at the top (‘Daleks’), and the fully-humanoid robots (bipedal, bibrachial, bimanual - legs/arms/hands) that in theory can cope with almost any terrain, including the outdoors (Cybermen).
Those humanoid robots (HuRos or huros) are what is lighting up the internet, particularly after demonstration events have shown what they are capable of: Football matches between teams of humanoids, some showing good ball skills and even tactics; A half marathon with 20 different humanoids competing (only six finished the 21km course, but with the winner finishing in a creditable 2:40); Boxing and Muay Thai matches between semi-autonomous humanoids; and even robots solving Rubik’s Cubes.
Perhaps it should come as no surprise that the majority of the companies involved in these events are Chinese, given that the sector is being heavily supported by the Chinese government. Forecasts for the market for huros suggest that it’s in the hundreds of millions of dollars now, but that the market will grow to multiple billions of dollars within a decade, largely driven by home help and domestic robots, those involved in providing care and companionship for our increasingly aged societies, and industrial huros.
The leading huros of today include the Tesla Optimus, Boston Dynamics Atlas, Unitree Robotics G1, 1X Neo and Agility Robotics’ Digit, but there are plenty of others1. All of them are being continuously improved, and the companies are learning from each other. Boston Dynamics has largely ditched its hydraulically-actuated huros, for all-electric models (like Tesla’s Optimus). Huros are adding many more degrees of freedom (‘joints’), with some into the 70s and beyond. This is impressive, but a human has more than 200 degrees of freedom. Huros increasingly have heavy lifting capabilities, currently up to around 50kg, allowing them to operate in industrial settings. Modular construction means that parts can be swapped out if required, like Wall-E getting a new eye. Many can change their own batteries, which is useful since battery capacity is one of the main limiting factors on their performance. They also feature advanced sensors to help them interact with objects and to navigate in their environments. Interestingly, some can communicate with their companion huros, so that if one learns a task, then the others learn it too.
The mechanics of huros will increasingly become a ‘solved’ problem: in the coming few years, advanced huros will be mass produced in robot factories (by humans - and robots), with economies of scale and severe (Chinese-led) competition bringing prices down to the US$30,000 level and lower. What will make all the difference, however, will be the installation of AI into these huros. They might be installed with general AI software, which can then learn specific tasks in local environments, with the software being upgraded as AI advances and while retaining its own learned knowledge. The huro is then merely the carcass - the AI is the brain. Interesting, but what’s it got to do with us?
Imagine an AI-driven huro that could build scaffolding, install refractories, clean silos, unblock ducts, undertake inspection rounds, tighten bolts, do the sweeping or vacuuming, and learn how to do all of the other dangerous or boring tasks that are required in a cement plant, for a one-off cost of US$30,000 (and a probable annual service contract). Spot, the robot dog at HM’s Leimen plant, has already shown the way forward for robotics in cement. Spot can’t make tea, but a huro can. Expect massive changes ahead.
1 https://humanoidroboticstechnology.com/articles/top-12-humanoid-robots-of-2026
