We explore the current leaders in humanoid robotics, their different approaches, the challenges facing the industry, and ask: 'when will we start seeing robots everywhere?'

It goes without saying that humanoid robots are going to be a big deal.

The estimated market size for humanoids ranges from $38 billion (Goldman Sachs) to $24 trillion (ARK Invest - I want to drink their Kool-Aid).

With the possibilities of automating millions of manual labor tasks and creating entirely new robo-butler markets, it is no surprise that the pie is estimated to be huge.

Fierce competition to build the best humanoid robot has certainly begun. Whoever produces the most effective humanoid robot deserves to take the largest piece of this ever-expanding pie.

Why humanoid?

What's so special about the human form that we need to build robots in the image of our feeble bodies?

Well, for two reasons:

1. Evolution appears to know what it's doing, and our bipedal design is fairly 'decent' for a long list of generalized activities.

2. The modern world is designed and purpose-built for our human form. Factories, offices, and homes are all designed with our dimensions and proportions in mind.

Key players in the race

Tesla Optimus

Tesla’s Optimus project, unveiled in 2021, is designed to leverage the company’s expertise in AI and mass manufacturing to build general-purpose humanoid robots capable of performing repetitive tasks in Tesla facilities and beyond.

They win both prizes for their name (Optimus is the coolest sounding out of all of the humanoid robots) and their aesthetic (the sleek proportions and white body look straight from a techno-optimist future… although they haven't solved the problem that the head can look like a scary faceless black void.)

Musk's approach, and advantage compared to many competitors, when it comes to deploying Tesla's humanoid, is that it can be 'dogfooded' on their internal production systems.

Recent reporting confirms that Tesla will produce a few thousand Optimus units by the end of 2025 (although Musk & Tesla timelines should be taken with a big fistful of salt), focusing on pilot lines in Fremont before full-scale rollout.

Figure AI

Figure AI has rapidly ascended the humanoid robotics ranks, securing $675 million in 2024 at a $2.6 billion valuation. There are now rumours that the startup is in the process of raising $1.5 billion at a $39.5 billion valuation. An insane increase from just last year. Is this the sign of a frothy bubble emerging? Or is it because Figure is making tangible improvements and amassing meaningful partnerships?

Figure has announced a prototype partnership with BMW to test production-line tasks, and is now in talks with United Parcel Service to integrate its robots into logistics operations (Yahoo Finance).

Figure also recently showcased its “Figure 02” autonomous humanoid, which promises a 5-hour runtime, 20 kg payload, and self-charging capabilities.

The matte black colour scheme certainly looks cool, but I can't help but see these robots as potential autonomous ninjas, and that does freak me out a little bit.

Unitree

Ah, yes, the Chinese robot manufacturer that boasts very impressive features (43 degrees of freedom, 2-hour battery life, 35kg weight) and equally impressive price points.

You can buy their G1 humanoid on the Unitree store right now, with prices starting at $16k. Their Go2 robot dog even retails for under $2k. The dogs look pretty damn terrifying, not least because they have been involved in some military training demos already (Daily Mail).

I can't help but see these robots as a blatant national security risk. Do we really want to be buying autonomous robots from China? They seem like the ultimate sleeper agent, ready to be remotely activated when the time comes. And many people are willingly paying to bring them into their house…

Boston Dynamics

No list about humanoid robots would be complete without mentioning the OGs: Boston Dynamics.

Their Atlas model remains the gold standard for dynamic humanoid robotics, showcasing backflips, parkour, and lifelike dance moves, albeit always in demo video setups.

Hyundai acquired Boston Dynamics in 2021, and it is reportedly bringing 'tens of thousands' of the Atlas & Spot models to its new Georgia 'Metaplant' factory (Newsweek).

A recent video highlighted Atlas performing tasks trained via reinforcement learning on human motion capture data, showing that the robot can be used in a dynamic environment and wasn't just full of pre-determined party tricks.

Agility Robotics

Agility Robotics' Digit deserves the award for cutest robot (and name).

Digit is a bipedal robot tailored for warehouse applications; much less humanoid than the others, with extended arms that help with handling larger loads.

Digit’s business model centers on scalable automation services, aiming to reduce operational costs and augment human labor in distribution centers. In April 2024, Agility announced a partnership with Manhattan Associates, a global leader in supply chain commerce.

Agility and Digit seem to be embodying the belief that purpose-built robots will ultimately win in important industries such as logistics. They aren't currently bothered with building a fun and helpful companion for consumers, they want to build an autonomous warehouse laborer that silently and diligently packs and unloads pallets 24/7.

Engineered Arts

Certainly the creepiest-looking robot on the list.

Engineered Arts’ Ameca is designed as a development platform for AI and interactive applications, featuring lifelike facial expressions, modular hardware, and an open SDK that enables rapid prototyping of social robotics use cases.

I imagine if (when) they give it a realistic skin tone, they could make an absolute killing selling this robot as the most realistic sex toy (for the sake of humanity, please don't do this).

And many more

This isn't, of course, an exhaustive list.

There are social robots like Hanson Robotics’ Sophia, other industrially-tailored robots such as Honda’s ASIMO, and SoftBank’s Pepper, to name a few.

The remote control scandal

As the race of the humanoid robots has heated up, each company tries to one-up the others by releasing bolder and more impressive demonstrations, through marketing videos and live events. Take Tesla's Cybercab showcase, where Optimus robots acted as the waiters and bartenders for the event.

However, the implicit claim that these robots were performing tasks such as navigating obstacle courses or serving cocktails autonomously is pretty misleading.

These recent demonstrations have frequently relied on human operators, hidden behind the scenes, leading to widespread criticism (rightly so).

Critics have likened such demos to Wizard-of-Oz experiments, where human puppeteering is passed off as advanced AI.

When will I get a robot?

Ignoring the hype from the manufacturers, most analysts still state they believe that safe, reliable, and versatile humanoid robots are at least a decade away.

However, Unitree's G2 sales prove that, in theory, robots are actually here (despite the risk that you are inviting a Chinese sleeper agent into your own).

It does feel like a game of 'fake it till you make it', taken too far?

With these manipulated demonstrations and misleading claims of capability, it is hard to truly pinpoint how far we have yet to go to achieve truly useful autonomous humanoid robots.

Sure, they can already be used as a fun toy with a repertoire of a few parlour tricks, but relying on them to do any chores inside the house? Not yet, unfortunately. If my Roomba still gets trapped in a corner of my kitchen, we still have some technical implementation details to solve.

Perhaps, unsurprisingly, the first place we will see Robots everywhere will be in the controlled environments of warehouse and factory floors. Agility Robotics is already being used, with Boston Dynamics, Tesla Optimus, and Figure AI allegedly closely behind.

Once the robots are released en masse to the factory floors, will this provide the necessary data to give them the skills and understanding needed to operate outside the factory and in our homes? How transferrable will the data collected and the learnings formed be?

My bet: 10,000+ robots deployed across factories by the end of 2025, 500,000+ by end of 2026, yet it takes 'until' 2035 before we start to see robots being adopted by the upper-middle class in the US and the rest of the world.

— Alex BB