What, exactly, are people expecting from domestic humanoid robots?

We could all use a helping hand. That’s not what you’re getting. 

Hugging Face recently bought Pollen Robotics, a startup developing “open source” humanoid robots. If that sounds weird, that’s because it is. The firm’s co-founder, Thomas Wolf, said in a post on X.com that the firm believes in a future where “AI and robots are open-source, transparent, and affordable; community-built and safe; hackable and fun.”

Ah yes, the holy grail of technology: making things that are simultaneously safe, hackable, and fun.

In the same post, Wolf declared that robotics was the “next frontier that AI will unlock.” We’re not trying to get into a semantics debate here but, arguably, robotics was the first field that AI unlocked. A significant percentage of all clothing and automobiles produced around the globe are manufactured with the use of autonomous machines. 

But, if we’re talking about a machine that just uses algorithms to perform a routine set of tasks, such as a mechanical arm that stitches hats, people don’t tend to call that artificial intelligence anymore. They want robots that look like humans, for some odd reason. 

Humanoid robotics

The robots Wolf specifically mentions in his post are the “Reachy 2” model from Pollen Robotics. You can buy your own for just $70,000. You’ll want to develop your teleoperator or Python-based control method however, because the point of the exercise is to roll-your-own robot. That’s what the “open source” part of the deal is all about.

Don’t get us wrong, we think this is a fantastic idea. It’s awesome that Hugging Face bought Pollen and the Reachy 2 is a really cool robot. But it’s meant for researchers, not the rest of us.

Reachy 2 isn’t going to make your life easier unless the challenges in your life involve needing a test platform to develop robotics solutions on. 

And that brings us to the core question: what, exactly, are humanoid robots supposed to do for the rest of us? Let’s try and imagine something useful for a robot to do in the average person’s life.

Maybe they could clean up

Right off the bat, we’re going to need cheaper robots. The average housekeeper in Los Angeles makes about $35K a year. We’ll need robots to either be much, much better than humans at all the tasks a housekeeper does, or they’ll need to be significantly less costly. 

Maybe they could do maintenance

Looking at Reachy 2, that thing looks like it could hold a screwdriver. But, before we put anything sharp in a machine’s hands, let’s just make sure it isn’t dangerous.

If we figure the amount of torque required for tightening a screw using the formula N = T / (2 * π * µ * d * p), where T is the torque, µ is the coefficient of friction, d is the diameter of the screw, and p is the pitch of the screw… and we compare that number to how much pressure it takes to break a human finger… 

It turns out that making robots that are strong enough to do chores requiring tools means making robots that are strong enough to break fingers. We won’t even get into hammer math. 

The only thing stopping the robot from accidentally or intentionally breaking things would be its programming. And not just parameters keeping it from experiencing an “instrumental convergence” problem where it determines that breaking things will help it determine its goals, but we’d also need to be sure that it never made a mistake when calculating the proper amount of pressure to apply for a given task. 

Which AI models never make mistakes? 

Maybe they could do easy tasks

It could hold trays of food for guests at parties, mix drinks behind a bar, and do the laundry. Of course, we could just set the food on a countertop. And, it seems like a waste to pay $70,000 for a robot bartender when you could program a cheap robot arm to do the same thing. And, anyone who can afford a robot already has a washer and dryer. That means you’d be paying for a complicated folding machine. 

Because you definitely don’t want a robot that’s strong enough to pull the laundry out of the machines and put it into a basket. A machine capable of carrying ten pounds of weight, while also being capable of manipulating objects as fine as cloth, and traversing carpets and tiles without falling over would need motors, servos, and enough weight to counterbalance that 10 pounds as it moved. 

That’s a machine that could cause some serious harm if it fell on a small child or pet. Which means it’s also a machine that pet owners, parents, and people with staircases in their homes might not want to have around.

Not the droids you’re looking for

We’re not saying that these machines can’t be safe. What we’re saying is that humanoid robots that are solid and powerful enough to move around your home and strong enough to perform most useful tasks are inherently dangerous. Any benefit a humanoid robot could provide, with the exception of “companionship,” could be better served by a dedicated machine.

Have you ever wondered why C3P0 looks like a person and R2D2 looks like a trashcan? It’s because Threepio is a “protocol Droid.” It’s supposed to be a diplomat, so it looks like most lifeforms it interacts with. Artoo, on the other hand, is an Astromech. Its job is, essentially, that of a pilot’s navigator. It was designed to be useful above all else. 

The trade-off between what humanoid domestic robots in the real world could be capable of and the potential for them to accidentally break your fancy dishes, step on your dog’s tail, fall on top of your coffee table, or get tangled up in a cord might not be what the average person is expecting. 

Ultimately, the people who are going to benefit the most from humanoid robots are those who’ll get their money’s worth: corporations that’ll use them as bellhops, receptionists, mascots, and moving mannequins to draw customers in. 

Read more: Putting LLMs inside of robots won’t solve the embodiment problem

Art by Nicole Greene