Before talking about humanoids, let's look at some facts about humans.
If you're not a program reading this, you can do so with eyes which see much better than the eyes of many animals — but are no match for the detail an eagle can make out.
If you listen to this text, your ears allow you to do so much better than the ears of a tortoise, but you're no match to your dog. And as we're talking about dogs: if they have a certain size, don't try to outrun them. You are not too bad at running, but still — many predators are faster than you.
The one thing humans do better than most animals is long-duration slow running — your dog might be the only animal which joins you on a marathon run (but please protect the paws).
Energy consumption is another topic which often gets overlooked. A human body operates on 2–3.5 kWh per day — a humanoid robot doing bodily work consumes that energy in one hour. And while the batteries need electrons to recharge, you can eat some fruit, vegetables or meat to recharge yours.
The human is, overall, a good and versatile compromise, fitting well into an existing ecosystem of resources. Our weird way of walking on two legs — which combines a high center of gravity with two fragile and unstable connections to the ground, requires permanent adjustment of balance and a complex concept of sensing movement and orientation, combined with immediate muscular reaction — enables us to do a number of different things in a mediocre way. We're the experts in doing many things "ok" and few things as the best.
So why do we want to re-engineer this type of machine and create humanoid robots?
I think the old dream of humans to create an artificial "human" which does the work we don't want to do is probably at the core of the movement. Being able to replicate human motions, human activity and having a machine doing human tasks is something we (as humanity) have been dreaming about since "ever" (have a look at Gods and Robots).
Probably the idea of having companions who simply do what you ask them to do is another attractive perspective — a humanoid won't complain if you ask them to bring out the trash or to clean up the kitchen (though we are still a few steps away from that level of accuracy).
Where's the evolution right now?
Creating mechanical solutions to make machines able to move like humans is a considerable effort. Our mechanics are the results of our own evolution, and trying to recreate how joints, limbs and ligaments work is possible — because we have a template — but takes time nonetheless. If you look at how robots are moving today, you come to realize that the human body is a quite impressive machine.
When we then discuss the use of humanoids in business, another question comes up: why make a compromise with a human-like, complex-to-navigate machine instead of creating machines which are built for one purpose and do that very well?
To my surprise, the answer comes from a humanoid robot manufacturer itself: in R&D, they manually assemble lot sizes of up to 10 in an unstructured, engineering-focused approach. Lot sizes from 10 to 100 are assembled by humans (!) in a manufacturing site designed for — humans. Lot sizes of 100 and above are outsourced into an automated production facility, where dedicated machinery (not humanoids) will assemble the humanoids.
Replace humans with humanoids in that scenario and you have the answer to where humanoids are the right fit: when you need flexibility and do not need to scale. If you've ever looked at industrial production, you know that there are many places in which production processes are not integrated and a "human interface" is transporting components from process step to process step — either because the process is not executed often enough to justify a real integration (like a conveyor belt or a robot arm), or because the step requires complex handling or movements and would be too expensive to automate.
So what we are doing is teaching machines to stand upright, to perform human tasks, to talk. And we're getting better and better, extremely fast. Today's humanoid robots move more smoothly with every software and hardware upgrade, get more run time out of their batteries and can use AI to react to unforeseen situations.
They will — and have to — become faster, but that's only a matter of time. Still, a purpose-built machine using modern technologies will always be faster when it comes to scaling. That said, humanoid robot development is key to advancing the technology of robotics and the combination of robotics and AI.
And in cases where you want to start automating things you could not automate before — because of complexity or low volume — humanoid robots can be the path to get started.
The other big advantage of humanoids is that they democratize automation. Right now, "localizing" robots requires specialist involvement. In a few years, this learning of layouts and processes will be powered by AI and make robotics an accessible technology for a much larger set of companies.
Humanoid robots are a segway into a new era of automation and will stay the tool of choice in the niche of innovation, flexibility and low-volume production.
If you want to learn more about this technology and develop your own robotics use case, join our Embodied AI Jam in Switzerland in August.
Bluesky
