Robots made The stage debuted on New Year's Day in 1921. More than half a century before the world got its first look at George Lucas's robots, a small army of silvery humanoid robots took to the stages of the first Czechoslovak Republic. They were, for all intents and purposes, humanoids: two arms, two legs, a head—the whole shebang.
Karel Čapek's play, RUR (Rossumovi Univerzální Roboti), was a great success. It has been translated into dozens of languages and played across Europe and North America. However, the lasting legacy of this work was its introduction of the word “robot.” The meaning of the term has evolved a bit in the past century, as Chapek's robots have been more organic than machines.
However, decades of science fiction have ensured that the public image of robots has not deviated too far from their origins. For many, the human form is still the platonic robot ideal, but the state of technology has not been able to catch up with this vision. Earlier this week, Nvidia held its own robotics demonstration on stage at its GTC developer conference, where CEO Jensen Huang was surrounded by images of six humanoid robots.
While the idea of a general-purpose humanoid concept has, in essence, been around longer than the word “robot,” until recently, the realization of this concept has seemed quite elusive. We're not there yet, but for the first time, this concept is on the horizon.
What is a “general purpose human being?”
Image credits: Nvidia
Before we go deeper, let's get two main definitions. When we talk about “general-purpose human beings,” the truth is that both terms mean different things to different people. In conversations, most people take a fair “I know it when I see it” approach to the conversation.
For the sake of this article, I will define a general-purpose robot as one that can quickly acquire skills and perform any task a human can do. One big sticking point here is that general-purpose robots don't suddenly become general-purpose overnight.
Because it is a gradual process, it is difficult to say precisely when the system has crossed that threshold. There's a temptation to go down a bit of a philosophical rabbit hole with that last piece, but in the interest of keeping this article within book length, I'll go ahead and move on to the other term.
I've received a bit of (largely mild) criticism when I've referred to the Reflex Robotics system as a human. People pointed out the obvious fact that the robot has no legs. Leaving aside for a moment that not all humans have legs, I would be fine with calling the system “human” or more specifically “human on wheels.” In my estimation, it resembles the human form enough to fit the bill.
A while ago, someone at Agility objected when I described Digit as “arguably human,” suggesting that there was nothing to argue about. What is clear is that the robot is not as sincere in trying to recreate the human form as some competitors. However, I will admit that I may be somewhat biased having traced the robot's evolution from its predecessor Cassie, who looked a lot like a headless ostrich (listen, we've all been through an awkward period).
Another element I tend to take into consideration is the degree to which the human form is used to perform human-like tasks. This element is not absolutely necessary, but it is an important part of the humanoid robotics ethos. After all, form factor proponents will be quick to point to the fact that we've built our worlds around humans, so it makes sense to build human-like robots to operate in that world.
Adaptability is another key point used to defend the diffusion of bipedal hominids. Robots have filled jobs in factories for decades, and the vast majority of them are single-purpose. This means that they are built to do one thing very well more often than not. This is why automation has been so well-suited to manufacturing – there is a lot of standardization and redundancy, especially in the world of assembly lines.
Brownfield vs. Greenfield
Image credits: Brian Heater
The terms “green field” and “brown field” have been in common use for several decades in various disciplines. The first is the larger of the two, and describes undeveloped land (literally, a green field). Developed to contrast with the previous term, brownfield refers to development on existing sites. In the world of repositories, this is the difference between building something from scratch or working with something that already exists.
There are pros and cons to both. Greenfields are generally more time and cost-effective, because they do not require starting from scratch, while greenfields provide the opportunity to build an entire site to specifications. Given unlimited resources, most companies will choose new domains. Imagine the performance of a space built from scratch with automated systems in mind. This is a pipe dream for most regulators, so when it comes time to automate, the majority of companies look for off-the-shelf solutions – doubly so when they first dip their toes into the robotic waters.
Given that most warehouses are brownfields, it is not surprising that the same is true for robots designed for these spaces. Hominins fit well into this category and, in fact, are among the most structurally resolved in a number of respects. This goes back to the previous point about humanoid robots constructing their environments. You can safely assume that most prefabricated factories were designed with human workers in mind. This often comes with items like ladders, which present an obstacle for wheeled robots. The size of this hurdle ultimately depends on a lot of factors, including planning and workflow.
Baby steps
Image credits: Figure
Call me a wet blanket, but I'm a big fan of setting realistic expectations. I've been doing this job for a long time and have survived my share of hype cycles. There is a extent to which they can be useful, in terms of building investor and customer interest, but it is all too easy to fall prey to over-promising. This includes stated promises about future functionality and demo videos.
I wrote about the latter last month in a post cheekily titled, “How to Fake a Robotics Demo for Fun and Profit.” There are several ways to do this, including hidden remote operation and creative editing. I heard whispers that some companies are speeding up videos without revealing the information. In fact, this is the origin of the name of the robotics company 1X – all of their demos run at 1X speed.
Most in the industry agree that disclosure is important—even necessary—with respect to such products, but there are no strict standards in place. One could argue that you are wading into a legal gray area if these videos play a role in convincing investors to pump in large sums of money. At the very least, they set wildly unrealistic expectations among the general public – especially those who tend to take the words of CEOs spreading the truth as gospel.
This can only do a disservice to those who are putting in the hard work while actually working with the rest of us. It is easy to see how hope quickly diminishes when systems fail to live up to those expectations.
The real-world deployment schedule has two primary constraints. The first is mechatronics: what devices can do. The second is software and artificial intelligence. Without getting into a philosophical debate about what can be described as artificial general intelligence (AGI) in robotics, one thing we can say for sure is that progress has and will continue to be incremental.
As Huang noted at GTC last week, “If we define AGI as something very specific, or a set of tests where a program can perform very well — or maybe 8% better than most — I think we'll get there.” Within five years.” Years.” This is the optimistic end of the timeline I've heard from most experts in the field. Five to 10 years seems to be common.
Before we reach anything resembling artificial general intelligence, humanoid robots will start out as single-purpose systems, like their traditional counterparts. Pilot programs are designed to prove that these systems can do one thing well at scale before moving on to the next. Most people look to carrying things for that low-hanging fruit. Of course, your average Kiva/Locus AMR can move bags all day long, but these systems lack the mobile handling tools needed to move loads in and out themselves. This is where robotic arms and end effects come in, whether they're attached to something that looks human or not.
Speaking to me a few days ago at the MODEX Expo in Atlanta, Robert Sun, Dexterity's founding engineer, brought up an interesting point: Humans could provide a clever stopgap on the way to powering down (fully automated) warehouses and factories. Once you have full automation in place, you won't necessarily need the flexibility of a robot. But can we reasonably expect these systems to be fully operational in time?
“By moving all logistics and warehousing work to automation, I thought humanoid robots could be a good turning point,” Sun said. “Now we don't have the human, so we'll put the robot there. Eventually, we'll move to this automated lights-out plant. Then the issue of human beings is very difficult which makes it difficult to put them in the transition period.”
Take me to the pilot
Image credits: Aptronic/Mercedes
The current state of humanoid robots can be summed up in one word: pilot. It's an important landmark, but it doesn't necessarily tell us everything. Test announcements arrive as press releases announcing the early stage of a potential partnership. Both parties love them.
For a startup, it represents real, demonstrable interest. For large companies, it signals to shareholders that the company is dealing with cutting-edge technology. However, real numbers are rarely mentioned. These usually come into the picture when we start discussing purchase orders (and even then, it doesn't happen very often).
Last year witnessed the announcement of a number of these matters. BMW is working with the shape, while Mercedes has enlisted Apptronik. Once again, Agility has a head start on the rest, having completed its pilot programs with Amazon – however, we're still waiting for word on what's next. It is particularly interesting that – despite the long-standing promise of general-purpose systems, almost everyone in this field starts out with the same basic functionality.
Two feet to stand on
Image credits: Brian Heater
At this point, the clearest path to general artificial intelligence should sound familiar to anyone with a smartphone. Boston Dynamics' Spot deployment provides a clear, real-world example of how the app store model works with industrial robots. Although there is a lot of compelling work being done in the world of robotics learning, we are a long way from systems that can discover new tasks and correct errors quickly and at scale. If only robot manufacturers could leverage third-party developers in a similar way to phone makers.
Interest in this category has increased dramatically in recent months, but on a personal level, the needle hasn't moved much in either direction for me since late last year. We've seen some absolutely killer demos, and generative AI offers a promising future. OpenAI is certainly hedging its bets, investing first in 1X and – more recently – in Figure.
A lot of smart people trust the form factor and a lot of other people remain skeptical. However, one thing I am confident in saying, is that whether or not future factories will be filled with humanoid robots on a large scale, all this work will amount to something. Even the most skeptical roboticists I've spoken to about this topic have pointed to NASA's example, where the race to tune into humans' moods led to the invention of the products we use on Earth to this day.
We will see continued breakthroughs in machine learning, mobile manipulation, and motion (among other things) that will impact the role automation plays in our daily lives in one way or another.
