Humanoid robots are coming. Ultimately?

That is The Stepback, a weekly publication breaking down one important story from the tech world. For extra on robotics and AI, comply with Robert Hart. The Stepback arrives in our subscribers’ inboxes at 8AM ET. Choose in for The Stepback here.

We’re taking a winter break! The Stepback might be again on January eleventh, 2026. In the meantime, you compensate for previous points here.

I’ve a gentle spot for robotic fail movies. I watch them on a loop, chuckling to myself, as a form of remedy. Perhaps I’m a sadist, perhaps I must get out extra — you’ll be able to choose me later — however they get me each time. So naturally, I’ve been glued to a clip of Tesla’s Optimus robot falling like a felled tree on the firm’s Autonomy Visualized occasion in Miami that’s been doing the rounds on social media this week.

In line with the footage, Elon Musk’s vaunted humanoid robotic was handing out water bottles from behind a desk earlier than knocking a bunch over, flailing its arms upward, and collapsing backwards like a puppet whose strings had been severed. Watch intently and also you’ll see two issues: a small plume of water as its arm crushes a bottle mid-fall (it made me giggle), and a movement uncannily much like somebody eradicating a VR headset.

It wouldn’t be the primary time Tesla has faked the autonomous part of its autonomous robotic, which Musk says is central to the corporate’s future. An early demonstration was only a dancer in a skintight bodysuit to point out what the Tesla Bot, now Optimus, might be. Later demonstrations had been revealed to be (moderately clearly, by the sound of it) humans in disguise, working the robots remotely with one thing like a VR headset, which we all know Tesla uses in development.

People have been obsessed with robots for centuries; from historic tales of stone golems and synthetic automatons to modern science fiction, driverless cars, and Roombas, the thought of animating the inanimate with one thing resembling life has fascinated us. A lot of the present hype for humanoids might be traced on to Musk, so it’s cheap to be greater than slightly bit skeptical when he and others promise they’ll revolutionize the world. Musk, who has vowed to build a “robot army” of 1 million humanoids, has a deserved fame as an outlandish and unreliable forecaster, and robotics has ridden quite a lot of hype waves in its historical past. Prior to now, know-how has at all times lagged the keenness of these desirous to deliver AI into the actual world, however right this moment, we’re being instructed that tech is lastly able to ship.

So what does able to ship truly appear like in 2025?

There’s positively a gold rush of types for humanoid robots proper now. Each main tech firm has them on their roadmap in some capability, and the likes of Nvidia, Meta, SoftBank, Google, Amazon, Microsoft, Intel, and Tesla (duh) are all throwing critical weight — and money — behind them as the subsequent large frontier in tech. And so they’re not alone: A rising constellation of challengers need in on the motion, together with Apptronik, Boston Dynamics, Determine AI, and 1X.

China desires in, too. Beijing has determined embodied AI — which additionally contains non-humanoids like drones, quadrupeds, and different autonomous machines — is the important thing to future financial development. In its trademark trend, it has moved to place itself as a world chief in robotics by huge funding, authorities directives, and state subsidies. Everybody from tech giants Ant Group and Baidu to startups like Unitree and AgiBot is piling in.

When you go by the demos China and everybody else is pumping out, you’d be forgiven for considering the humanoid robotic future has already arrived. This summer season, robots competed in dance, fight, monitor, and subject occasions on the first World Humanoid Robot Games in China. An identical occasion was held in Greece — the International Humanoid Olympiad — on the birthplace of the Olympic Video games. Humanoid fights are extra standard than I might have imagined and are apparently going down completely all over the place, from organized contests and underground fight clubs to sparring with CEOs.

Corporations are additionally eager to get bots out of factories and into the house. These sorts of human-centric areas are why advocates say humanoids are price pursuing as a substitute of different robotic physique varieties that could possibly be simpler to provide. Determine mentioned its new Determine 03 robotic can tackle home chores, releasing a video of the bot doing dishes and folding garments. 1X debuted Neo, which it claims is protected and “the world’s first consumer-ready humanoid robotic.” There are additionally videos of Neo shakily finishing primary family duties. I’m unsure I can afford it, however for those who’re , 1X is promoting models for $20,000, with supply beginning within the US subsequent yr.

However precise use instances stay slim, and spectacular demos usually are not the identical as working merchandise. For all of the progress — and there have been spectacular advances — demos are staged, scripted, and even teleoperated. For instance, whereas it’s technically true to say that Ant Group’s R1 was cooking for audiences at a commerce present this yr, it moved at a tempo so glacial it might’ve thrilled Meryl Streep’s Miranda Priestly. Equally, the thought of getting 1X’s Neo sounds a lot much less interesting when you notice you have to additionally decide to having random humans telecommute into your home to manage it remotely. And likewise, robotic sport occasions are so enjoyable to observe as a result of the contestants are wobbly, unsure, and unpredictable, not as a result of they’re proficient athletes.

All of this raises the apparent query: If the tech clearly doesn’t match the hype, why are so many individuals getting so enthusiastic about humanoids? Properly, for the primary time in a very long time, it’s not that loopy to assume the tech could be beginning to catch up.

Traditionally, it has been exceptionally troublesome to get robots to carry out even mundane duties that people do with out considering, like strolling or choosing up a glass of water. {Hardware} — however principally software program — was a limiting issue, and machines had been confined to extremely structured environments and really slender duties. Advances in AI are altering that and turbocharging robotic improvement.

The extraordinary breakthroughs in AI picture and textual content technology from firms like OpenAI and Google are largely right down to massive language fashions, or LLMs, that energy chatbots. These work by utilizing complicated networks to acknowledge and reproduce patterns discovered within the huge quantities of knowledge they’re skilled on. Larger AI fashions are skilled on virtually all the things that may be scraped from the web, and the ensuing techniques are able to performing in a generalized means, moderately than following inflexible, hand-coded guidelines. To perform in the actual world, robots want this sort of generalization, and roboticists are attempting to make use of the identical strategy to offer machines the versatile understanding of the bodily world they’d must survive in it.

The problem right here is information. The form of materials used to coach LLMs was abundantly obtainable on-line, however the information wanted to coach robots isn’t practically as simple to return by. Robots want detailed examples of issues shifting in the actual world, which doesn’t actually exist at scale. Corporations are working onerous to vary that, and there’s a large effort to create the form of information wanted to coach robotic fashions at scale. On the floor, this seems to be ridiculous — Tesla has employees don cameras and sensors to show Optimus to behave like a human — nevertheless it’s working. It’s additionally why firms like 1X are deploying less-than-autonomous bots into peoples’ houses, as doing so gives them the opportunity to assemble the info wanted to make autonomous bots by conducting complicated duties remotely, comparable to loading the dishwasher or cleansing a tough object.

Humanoids have gotten extra inexpensive, too — particularly in China — as {hardware} prices sink and economies of scale kick in. Whereas costs range wildly — entry-level models might be as little as $1,400 for Chinese model Bumi to round $13,500–$20,000 for fashions from Unitree or 1X, whereas industrial fashions can value greater than homes — they’re approaching costs inexpensive by some shoppers, which means extra bots are being deployed in the actual world. This creates a suggestions loop, as firms get extra information to work with, they construct higher fashions and higher robots that extra individuals would need, presuming they’re keen to welcome them into their houses.

However even with all this progress, it might nonetheless be hype. In November, China’s main financial planning company warned a humanoid robot bubble could be brewing, contrasting the sheer variety of firms and scale of funding with the absence of viable use instances. The actual fact is, they’re probably not autonomous but, so, apart from hobbyists and researchers, who of their proper thoughts would purchase one? If I wished to get somebody in to wash my residence, I might rent a cleaner with out the hefty upfront value and save my cash till somebody produces a robotic that may truly do the duty I purchased it for.

Till firms cease hiding behind glitzy promo movies and teleoperating purportedly autonomous robots, it’s going to be onerous to inform the place we’re. Perhaps the robots are lastly coming, or perhaps I’ve many extra fail movies coming my means. Time will inform, I assume, however I’ll get the popcorn prepared.