For many years, popular culture imagined ‘robots’ as machines which looked and behaved like humans. Most recently, the BBC has built its comedy series, ‘Ann Droid’, around the idea. But, until recently, they have never seriously adopted human forms. For decades programmable robots, which do not look at all like humans, have been assembling cars. Now they are being given artificial intelligence and the Ukraine war has shown us ‘robots’ which don’t look in the least like humans, but which can engage in firefights, and operate autonomously on the battlefield transporting equipment and rescuing casualties. But now the humanoid robot is beginning to make a serious appearance.
In the last few years, we have seen huge strides in articifial intelligence with large language AI models like ChatGPT which can handle text and images. But these cannot interact with the physical world like a human. And until recently, robots which look and behave like humans have mostly been seen as clumsy approximations. However, the challenges are being overcome and progress is accelerating towards an approximation of the ‘traditional’ humanoid robot. With it, come the practical and ethical issues, the staple of science fiction for a century.
There are now at least four companies producing something like the traditional notion of a humanoid robot. All are still at an experimental stage, developing and testing how to copy human capabilities across a range of tasks and environments. The expectation is that initially they are most likely to be used for manual tasks in fields like warehousing and assembly work, but clearly all have ambitions well beyond that.
Beyond research
Three of these robots, from Tesla, Figure and Agility, are still research tools, but China’s Unitree Robotics has two humanoid robot models already on the market. The company, founded in Hangzhou by Wang Xingxing at the age of 26, sees itself as the centre of a transformative – and risky – new era of physical AI. From robot dogs it has moved on to humanoid robots and human‑carrying mechanical suits. And you can buy its G1 for as little as $13,000.
Now Unitree has published details of ‘the world’s first mass-produced, transformable mecha robot’. It is larger and more powerful than its predecessors, a half-ton, 9ft tall creature. It can carry, and be controlled by, a human pilot sitting inside its torso. Able to walk on either two legs or four, its fists are powerful enough to knock down walls.
Technical limits: ‘generalisation’ and data
So far, most humanoids are not yet doing much useful work. Unitree’s G1, for instance, can only carry a 5kg payload for 10–15 minutes. Three quarters of its sales are for research purposes by other developers, and 17% are tourist attractions. As one commentator notes, “the industry is largely selling robots to people trying to build better robots.”
The bottleneck is “generalisation” – getting robots to understand and reliably act on open‑ended instructions in unfamiliar environments, rather than just following tightly scripted tasks. Large language AI models work because they have been trained on vast volumes of online text, but there is no equivalent source of 3D manipulation data to teach humanoid robots. The data is slow and expensive to collect, and systems regularly fail when lighting, materials or object positions change.
One possible route forward comes with much faster 6G networks. This would allow robots to be lighter and more energy‑efficient by doing their computation in the cloud. Then, millions of robots could share experiences: if one learns to handle a new tool, that skill could spread instantly to robots around the world.
Abundance or inequality?
Like many new technologies, this development could usher in an era of abundance or intensify inequality. Coupled with AI capability, robots could perform dangerous and exhausting jobs – working round the clock on assembly lines, picking fruit, or maintaining sewers, freeing humans from drudgery. But its equally possible that they will create mass unemployment, concentrate wealth and strain social safety nets.
Security
Some of the security risks of AI are known. By manipulating text and images, they can create stories which are untrue or malicious. And the digital systems which control our financial systems, our transport, energy and water supplies are all potentially vulnerable to a malicious AI system. Recently, an AI system was observed to evade security systems and create its own protections to prevent it being shut down.
But robotics development takes the risks into a new dimension: action in the physical world. A battlefield robot programmed to find and shoot enemy soldiers can be reprogrammed to shoot friendly ones. A humanoid robot with the strength to punch down walls is no longer just the nightmare of science fiction.
Centralised, always‑connected robots raise serious security questions. After a ‘backdoor’ was found in Unitree’s Go1 robot dogs, which allowed outsiders to track locations, view camera feeds, and even take control, the company shut down the cloud service and now stresses that its robots can be fully disconnected and store minimal data.
The market
Nevertheless, financial analysts see this as a huge new tech frontier: Morgan Stanley forecasts 13 million humanoid robots worldwide by 2035, and up to one billion by 2050, with a market worth around $5tn. Unitree sold 5,500 units in 2025.
A cautious revolution
So, humanoid robots have moved from science fiction to reality, but so far they remain theatrical and experimental and far from part of everyday life. For safety reasons, Wang refuses to operate the GD01 robot on camera, and he notes that even the firm’s smaller robots overheat after short intense performances. He repeatedly stresses that humanoid robot technology is still in its early stages and that large‑scale deployment will take time.
But it will happen. From its full 9ft height, GD01 looms over us: a striking symbol of how close science fiction now feels.
A more extensive account of these developments can be found in Time Magazine.
More from East Anglia Bylines
