Understanding Humanoid Robots

[ad_1]

robots built them The stage opened the day after New Year 1921. More than half a century before the world caught its first glimpse of George Lucas's droids, a small army of silvery humanoids arrived at the feet of the First Czechoslovak Republic. They were, for all intents and purposes, humanoid: two arms, two legs, a head – the whole shebang.

Karel Čapek's play, RUR (Rosumovi Univerzalny Roboti), was a hit. It was translated into dozens of languages ​​and played throughout Europe and North America. However, the lasting legacy of the work was the introduction of the word “robot”. The meaning of the term evolved considerably in the intervening century, as Capek's robots were more organic than machines.

However, decades of science fiction have ensured that the public image of robots has not strayed too far from its origins. For many people, the humanoid form is still the ideal of the Platonic robot – it's just that the state of technology has not yet reached that vision. Earlier this week, Nvidia held its own robot parade on stage at its GTC developer conference, where half a dozen humanoids were photographed alongside CEO Jensen Huang.

While the notion of the concept of a general-purpose humanoid has, in essence, been around much longer than the word “robot”, until recently, the realization of this concept seemed completely incomprehensible. We're not there yet, but for the first time, the concept appears to be on the horizon.

What is a “general-purpose humanoid”?

NVIDIA CEO Jensen Huang seen presenting at GTC 2024

Image Credit: NVIDIA

Before we delve deeper, let's explore two key definitions. When we talk about “general-purpose humanoids”, the fact is that both terms mean different things to different people. In conversation, most people take Justice Potter's approach of “I know it when I see it”.

For this article, I'm going to define a general-purpose robot as one that can quickly learn skills and perform essentially any task that a human can do. A big problem here is that multi-purpose robots don't suddenly become general-purpose robots overnight.

Because it is a gradual process, it is difficult to say exactly when a system has crossed that limit. There's a temptation to go a little down the philosophical rabbit hole with that latter part, but to keep this article under book length, I'm going to go ahead and skip to the second term.

I got a little (largely good-natured) backlash when I referred to Reflex Robotics' systems as humanoid. People pointed out the obvious fact that the robot has no legs. Putting aside for a moment the fact that not all humans have legs, I have no objection to calling this system “humanoid” or more specifically “wheeled humanoid.” In my estimation, it looks similar enough to the human figure to fit the bill.

A while ago, someone at Agility took issue when I called Digit “arguably a humanoid”, suggesting there was nothing debatable about it. It's clear that the robot is not as convincing an attempt to recreate the human form as some of the competition. However, I will admit that I may be somewhat biased in tracking the robot's evolution from its predecessor Cassie, which resembled more than a headless ostrich (listen, we all went through a weird phase).

Another element I consider is the degree to which the humanoid form is used to perform human-like tasks. This element is not absolutely necessary, but it is an important part of the feeling of a humanoid robot. After all, form factor proponents will be quick to point to the fact that we built our world around humans, so it makes sense to create human-like robots to work in that world.

Adaptability is another important point used to defend the deployment of bipedal humanoids. Robots have had factory jobs for decades now, and most of them are single-purpose. That is to say, they were built to do the same job very well multiple times. This is why automation is so well-suited for manufacturing – there's a lot of uniformity and repeatability, especially in the world of assembly lines.

Brownfield vs Greenfield

Agility issue at this year's Modex conference

Image Credit: brian heater

The terms “greenfield” and “brownfield” have been in common use for several decades in a variety of disciplines. The first is one of the two, describing undeveloped land (quite literally, a green field). Unlike the first term developed, brownfield refers to development on existing sites. In the world of warehouses, there is a difference between building something from scratch or working with something that already exists.

Both have advantages and disadvantages. Brownfields are generally more time and cost effective, as they do not require starting from scratch, whereas greenfields offer the opportunity to create a site completely custom-made. Given unlimited resources, most corporations will opt for greenfield. Imagine the performance of a space built from the ground-up with automated systems in mind. This is a pipe dream for most organizers, so when the time comes to automate, most companies look for brownfield solutions – doubly so if they're dipping their feet in the robotic waters for the first time.

Given that most warehouses are brownfield, it should come as no surprise that the same can be said for robots designed for these locations. Humanoids fit well into this category – in fact, in many ways, they are one of the grayest solutions. This comes back to the earlier point about creating humanoid robots for their environment. You can safely assume that most brownfield factories were designed with human workers in mind. It often comes with elements such as stairs, which create obstacles for wheeled robots. Ultimately how big that bottleneck is depends on a number of factors, including layout and workflow.

baby steps

Image Credit: Shape

Call me a wet blanket, but I'm a big fan of setting realistic expectations. I've been doing this a long time and have survived my share of hype cycles. They can be useful to a certain extent in terms of garnering interest from investors and customers, but it is entirely easy to fall victim to overpromise. It includes both promises regarding future functionality and demo videos.

I wrote about it last month in a cheekily titled post, “How to Build a Fake Robotics Demo for Fun and Profit.” There are several ways to do this, including hidden teleoperation and creative editing. I've heard whispers that some companies are increasing video speeds, without disclosing information. In fact, the name of humanoid firm 1X originates from this – all of their demos run at 1X speed.

Most people in the field agree that disclosure on such products is important – even required – but there are no strict standards. One could argue that if such videos play a role in convincing investors to plunk down huge sums then you are moving into a legally gray area. At the very least, they set wildly unrealistic expectations among the public – especially those inclined to take the words of truth-distributing authorities as gospel.

This can only serve to harm those who are in reality working hard to keep up with the rest of us. It's easy to see how hope quickly diminishes when the system fails to live up to those expectations.

There are two primary constraints on the real-world deployment timeline. The first is mechatronic: that is, what the hardware is capable of doing. The second is software and artificial intelligence. Without getting into philosophical debates about what qualifies as artificial general intelligence (AGI) in robots, one thing we can say for sure is that progress has been – and will continue to be – gradual.

As Huang said at GTC last week, “If we defined AGI as something very specific, a set of tests, where a software program could do very well — or maybe 8% better — than most people, So I believe we'll be there within five.” years.” I've heard from most experts in the field that this is on the optimistic end of the timeline. A five to 10 year period seems to be typical.

Before we hit anything like AGI, humanoids will start out as single-purpose systems like their more traditional counterparts. Pilots are designed to prove that these systems can do a job well at scale before moving to the next level. Most people are looking for the lowest hanging fruit in tote moving. Sure, your average Kiva/Locus AMR can move toys around all day, but those systems lack the mobile manipulators needed to get the payload on and off themselves. That's where robot arms and end effectors come in, whether they're attached to anything human-looking or not.

Speaking to me the other week at the Modex show in Atlanta, Dexterity founding engineer Robert Sun made an interesting point: Humanoids could provide a clever stopgap on the way to illuminating warehouses and factories. Once full automation is in place, you won't need the flexibility of the humanoid. But can we reasonably expect these systems to be fully operational on time?

“Converting all logistics and warehousing tasks to robotic tasks, I thought humanoids could be a good transition point,” Sun said. “We don't have the humanoid anymore, so we'll put the humanoid there. Eventually, we'll head to this automated lights-out factory. Then there is the issue of humanoids being very difficult to keep them in the transition period.

take me to the pilot

astra

Image Credit: apptronic (Opens in a new window)

Image Credit: Aptronic/Mercedes

The current state of humanoid robotics can be expressed in one word: pilot. This is an important milestone, but it doesn't necessarily tell us everything. The pilot announcements come in the form of press releases announcing the initial phase of a potential partnership. Both parties love him.

For startups, they represent genuine, proven interest. For a large corporation, they signal to shareholders that the company is engaged in cutting-edge operations. However, actual figures are rarely mentioned. These usually come up when we start discussing purchase orders (and even then, not often).

The past year saw many of these announcements. BMW is working with Figur, while Mercedes has enlisted Aptronic. Once again, after completing its pilots with Amazon, Agility has the edge over the rest – however, we are still awaiting information about the next phase. It's particularly telling that – despite the long-term promise of general purpose systems, almost everyone in the field is starting out with the same basic functionality.

two legs to stand on

Modex Toy Automation Factory Warehouse Robot

Image Credit: brian heater

At this point, the most obvious path to AGI should look familiar to anyone with a smartphone. Boston Dynamics' spot deployment provides a clear real-world example of how the app store model could work with industrial robots. Although there is a lot of compelling work being done in the world of robot learning, we are a long way from systems that can detect new tasks and correct mistakes on a large scale. If only robotics manufacturers could take advantage of third-party developers like phone makers do.

Interest in this category has grown substantially in recent months, but personally speaking, the needle hasn't moved much in either direction for me since late last year. We've seen some absolutely fantastic demos, and generative AI presents a promising future. OpenAI has certainly been hedging its bets, investing first in 1X and – more recently – Figure.

Many smart people believe in the form factor and many others remain skeptical. However, one thing I am confident in saying is that whether or not the factories of the future will be populated by humanoid robots on a meaningful scale, there will be something to do with all of these tasks. Even the most skeptical roboticists I've talked to on the subject have pointed to the NASA model, where the race to get humans in the mood led to the invention of products we use on Earth to this day. We do.

We are going to see continued breakthroughs in robotic learning, mobile manipulation, and locomotion (among others) that will impact the role of automation in our daily lives in some way or the other.