Based on Business Matters (bmmagazine.co.uk), reporting by The Next Web and the BBC, and Figure AI’s public demonstration (May 2026)
The UK’s Sharp Group has put a human-shaped robot named Alpha to work at its recycling plant in Rainham, east London. The story is almost tailor-made for a “robots are replacing people” headline: injury and ill-health in the sector run 45% above average, the fatality rate is a multiple of the national figure, and annual staff turnover sits at 40%. The work is dirty, dangerous, and nearly impossible to staff. The robot, as finance director Chelsea Sharp puts it, “stays where you put it, works around the clock, won’t ask for a holiday, and won’t call in sick.”
All of that is true. But to my mind, the most interesting question here isn’t about people, and it isn’t about the ethics of automation. It’s about form.
Alpha is a body built by China’s RealMan Robotics, adapted for recycling work by the British startup TeknTrash Robotics. And betting on a humanoid shape in this industry is unusual. The argument for it is honest and down-to-earth: TeknTrash founder Al Costa says a human-like form lets the robot slot into a plant’s existing layout without redesigning the place around it. In other words, the humanoid here is bought not for speed, but for compatibility with an old facility built for people.
And that’s where the real story begins — because on speed, a humanoid is far from the obvious choice.
The alternative — purpose-built sorting systems engineered for one task — has been developed for years by America’s AMP (Colorado) and Glacier (California), and AMP’s systems already exceed human sorting speed by a factor of eight to ten.
So what do the humanoids themselves show at their best? In May, Figure AI staged a public marathon: its F.03 robot against a human at sorting parcels (a warehouse, not a tip — gentler conditions). The result was honest and sobering: the human still won — 12,924 parcels to 12,732, at 2.79 seconds per item for the human against 2.83 for the robot. That is, one of the best humanoids in the world, backed by billions from Microsoft, Nvidia, and OpenAI, barely reached parity with a person after all the hype — and that was on a tidy warehouse stream, not a chaotic line of shoes, concrete, and old cassette tapes.
Figure’s marathon is genuinely impressive on a different metric — endurance: a fleet of robots ran for roughly 200 hours without a hardware failure, processing nearly 250,000 parcels. But that’s about reliability, not speed — and speed is what we care about here.
Put the picture together (carefully — these are three different tasks and environments, not a direct comparison): purpose-built automation is already several times faster than a human; the best humanoids are roughly at human level; and Alpha, on a dirty stream of mixed waste, is still only learning, with no published speed figures at all.
Hence my takeaway. Sharp’s bet on a humanoid isn’t a bet on productivity. It’s a bet that it’s easier to drop a human-shaped machine into an old plant than to rebuild the plant around a purpose-built line. For a family firm with a single facility, that may well be rational. But as an industrial trend, the humanoid-on-a-sorting-line will have to answer an awkward question: why pay for a universal human-shaped form where a task-specific machine already works faster and cheaper?
The most interesting thing in Rainham isn’t that a robot is stepping onto the line. It’s that it’s stepping on in a shape that may have been chosen for the wrong reason.