The Next Robotics Revolution - Matthew T. Mason

11 min read Original article ↗

SHARE:

A secret plan to return the field of robotics to the good old days, by disguising it as a revolution.

I am part of a secret cabal, the Society of Crotchety Old Roboticists, or SCORs. The SCORs are plotting to turn back the clock and return the field of robotics to the good old days. After a few meetings, we hit on a plan: a return to the good old days, disguised as the next robotics revolution. Here’s the plan, unanimously approved by the SCORs. Don’t tell anybody under the age of 60.

The SCORs’ plan is to declare a new robotics revolution, replacing the old humanoids vision with an exciting new vision. Why? As a start, let’s revisit the previous robotics revolution, which took us from factory automation to humanoids.

Many years ago, factory automation was the dominant application domain for research in robotic manipulation. Then a revolution swept the field, and factory automation became the tired old dream. According to this new way of thinking, factories were structured environments — modified to simplify the robots’ tasks and eliminate all uncertainty. Each robot was designed for just one narrowly-defined task. Factories were a solved problem. So they said.

What was their new vision? Humanoids. They would focus on unstructured environments, like your kitchen. The future would be general-purpose robots resembling humans and performing the wide variety of tasks that humans perform. Welcome to the age of humanoids.

We SCORs might have responded thus:

Factories a solved problem??? Balderdash! Uncertainty eliminated? Hogwash! Much of our research focuses on the inevitable uncertainty in factories. And do you really believe that your kitchen is unstructured? Laughable.

But who was going to listen to that, when the humanoids looked so great on YouTube? Whether you agreed with the old guys or not, what could you do? People were excited about the new direction. They had made their choice. They were not going back.

If we cannot go back, let’s go forward. Let’s take the offensive! Time to declare the next robotics revolution:

The next robotics revolution has arrived!

What’s next? We need to dismiss the old vision. Right now there is so much excitement about humanoids in the home. How do we address that? One way would be to declare the existing work a failure. Maybe we could take all their sped-up videos, and post them on YouTube running at actual speed. And, we could point at their robots’ published success rates, which can be quite low.

But a better way to dismiss the humanoids vision would be to declare it a success. Kitchens, like factories, are a solved problem. Just look at those awesome videos! And for anybody who doubts, notice that the torch is passing from university research groups to industrial groups that promise products in the next couple of years. So, you have succeeded fabulously. Well done! Congratulations! Time to move on.

Second, we need some catchy labels. When people attacked the old vision in AI, they coined the term GOFAI — Good Old Fashioned AI. So, Good Old-Fashioned Robotics? Or, perhaps, Humanoia? Or, if we called the new era New-Fangled Robotics, then we could call the previous era Old-Fangled Robotics. And we could label the people as well: NewFers versus OFers. Pronounced so as to rhyme with loafers. Or, as in o-for-one.

What else is missing? Substance. What is the actual substance of New- Fangled Robotics? We have dismissed the tired old vision. What is the exciting new vision?

Here’s an idea. If you ask what would be the natural sequel to humanoidism, it might be to generalize. Instead of a narrow focus on robotic humans, we embrace all of evolution’s animal designs — the whole menagerie. Robotic chimpanzees (chimpanzeeoids?) could scamper through our warehouses, up and down those ridiculously tall shelves, tossing items down to their comrades below. Robotic rats (ratanoids?) could run around on the floor at your house cleaning up and retrieving dropped items. What about dealing with pests? A humanoid is going to be no better at dealing with mosquitoes or cockroaches than a human. Maybe some robotic frogs for the mosquitoes. Maybe a squad of robotic cockroaches that could pursue nature’s cockroaches under the baseboards and dispatch them.

Copying animals is one possibility, but we could go further and copy the evolutionary process itself, as viewed by ecologists. Let’s call that a second phase. We choose habitats of interest, and then, for each habitat, we would define “ecological niches” and develop new robotic creatures for each niche.

Which habitats? We could revisit the home. The OFers focused on just one ecological niche, occupied by their humanoids. But in nature every habitat contains several ecological niches. What other niches should we design for the home? We already have one: the niche long occupied by robotic vacuums like the Roomba. Humanoids might be well-suited to baby-sitting, but for dog-sitting? And, is the chef’s niche really the same as the upstairs maid’s niche?

And then, what other habitats? Warehouses, naturally. And … factories! We have come full circle! Factory automation redux!

Here is a first attempt at a manifesto.

The New-Fangled Robotics Manifesto

Robotics has passed through many stages: behavioral, minimalist, embodiment, probabilistic, developmental, and others. Continued progress depends on a constant recycling of old ideas under new labels. Wait, no, I mean progress depends on constant re-examination of past assumptions, and exploration of new ideas.

Most recently the field has devoted generous resources to the development of general-purpose anthropomorphic robots for domestic service — humanoids. We are encouraged by the stunning success of humanoids, and ready to begin the exploration of new ideas. The goal of this manifesto is to gather a few new ideas under the moniker New-Fangled Robotics.

Phase I: Animaloid Robotics

New-Fangled Robotics consists of two phases, which might occur sequentially or in parallel. The first phase is Animaloid Robotics. Drawing on the success of humanoid robotics, we extend our work to the other life forms devised by evolution. Humans might be the best design for the domicile, if we are constrained to a single design. But other forms could also contribute. A Dogoid could be a good companion for the household pet dog, while secretly reporting any shenanigans to the human master. ARatoid could help keep the floor clean, and retrieve dropped objects. Ratoids could also be useful in the workplace. Chimpanzeeoids would be great in warehouses, scampering up and down the shelves to retrieve items, and tossing them down to humanoids who pack the orders.

For each animal we copy, we might want to design a variant: a robotic videographer to produce stunning videos for YouTube. Imagine the spectacle of robot cockroaches battling natural cockroaches inside your walls. Certain to draw millions of views.

Phase II: Niche Robotics

The second phase is Niche Robotics. Niche Robotics does not copy natural evolution’s designs, i.e., animals. Nor does it copy the processes by which evolution designs the animals. Instead, we are inspired by the language used in ecological studies. First, we choose a habitat. Then, for each habitat, we design ecological niches and corresponding robots, using all available engineering tools.

Some habitats of interest:

  • The home. Besides the kitchen chef niche and the vacuuming niche, we might want to look at a pest control niche, an elder care niche, or many others.

  • Warehouses. Again some niches have been identified and filled, but warehouses are complex places with many more niches to define and fill.

  • Factories. A similar story here. Many niches remain for us to identify and fill with robots.

And there are many more: operating rooms, agricultural fields, planets to explore … on and on.

Welcome to the future of robotics! Just like the past, but newer!

Inline Feedbacks

View all comments

Virus-oids. They track down bad viruses. They can take their evolutionary clues from another failed technological miracle – vaccines.

How about bore-oids? They take up our time when we are bored and amuse us. They would replace another failed technological mirecle – the UTube channel.

I misread animaloids as amyloids. Perhaps that’s fitting for a charter member of SCOR.

Ratoids could be awesome. Imagine the objective of Ratoid is to figure out the rat tunnel in a field based on random rats that appear, learn rat language, rule all rat and guide all rats to a rat world war and destroy every rat ☠️.

Reply to  winner

22 days ago

Hahaha! Not sure whether everybody is seeing an enormous skull and crossbones, but there is a bug somewhere that causes some icons to render large. Very effective in this instance.

But, the concept! It’s brilliant. It’s the Pied Piper fairy tale, but plausibly implemented with modern technology. Another existential danger of AI, if you’re a rat.

Last edited 22 days ago by Matt

Sciuroids to clean up powerlines from the brunches that keep causing power outages (and may be learn also some electromagnetism to repair them!)

Last edited 22 days ago by Kostas Daniilidis

The humanoids you are describing are just anthropomorphic bodies (unless they are teleoperated in which case the brain is also anthropomorphic…). The future of robotics will be similar the future of autonomous cars: “understanding”. However somersaulting has been solved…

I don’t know about you, but I for one would like to become an eternal cyborg in the future.

“Chimpanzeeoids would be great in warehouses, scampering up and down the shelves to retrieve items, and tossing them down to humanoids who pack the orders.” –> I like this! So funny.
I need a Cat-oid to keep one cat from stealing another cat’s food…

Hey Matt, huge fan of your work – few questions (which are also framed somewhat like answers/loaded with opinions)

1.   On form factor origins: Did early roboticists converge on the articulated arm due to genuine task optimality, or path dependence from existing industrial machinery? Put differently: if we were designing factory robots from scratch today, would we still arrive at the 6-DOF arm?
2.   On the demo-to-ROI transition: When did factory robotics cross from ‘impressive demonstration’ to ‘obvious business case’ – and do you believe humanoids are currently stuck in the demo phase, or approaching that transition?
3.   On biomimicry as constraint: Your animaloid phase uses biological form factors as design priors. But evolution optimizes under biological constraints – metabolic cost, available materials, developmental pathways – that don’t bind engineering (if cost isn’t a concern). Should we expect optimal robotic designs to diverge radically from any biological precedent, or does evolution still provide useful starting points?

4.   On generality vs. niche: You implicitly argue that generalizability imposes a tax without sufficient benefit. But couldn’t a general-purpose platform achieve the economic advantages of shared R&D and manufacturing scale? Is ‘general-purpose’ perhaps better understood as ‘good enough across many niches’ rather than ‘optimal at one’ – and might that still win on unit economics?

Reply to  Chaitanya Sharma

8 days ago

Wow, lots of good stuff there.

  1. 6-DOF arms are not the rule, Sometimes they are favored for “generality” — able to move the attached effector in any direction — x,y,z,yaw,pitch,roll. But some articulated arms have only 5 DOFs, only two wrist motions. And then there are SCARAs and Deltas. There are also lower DOF devices, not always called robots because the RIA defined “robot” to mean three or more axes. Details are sometimes chosen to simplify the kinematic calculations. But I think your question is a good one. If you optimized for some set of tasks, you would probably get a different design.
  2. Most histories focus on the Unimate as the first successful industrial robot, though I have read of others of similar vintage. Are humanoids stuck in the demo phase? “Humanoids” is a broad term. There is a big difference between the most visionary visions, and the more practical visions. So, it is hard to describe or speculate on where it is all headed. It is amazing, though, even disconcerting, to see such a dramatic commitment of commercial resources to an advanced research project.
  3. Good point, well said. Evolution optimizes under constraints that don’t bind engineering.
  4. Your argument for generality definitely applies at some level. We do not manufacture a different robot arm for every task. Nor do we imagine a single design for all tasks. The answer lies in between, but that middle ground is tricky. We cannot characterize the sets of tasks we might wish to address, not with precision, so we fall back on the vague phrase “whatever humans do.”

The sentence about recyling old ideas under new labels made me laugh out loud. I like to tell people that if the robot has a 95% success rate, that means it’s dropping on average one dish every time it loads your dishwasher. Won’t take much of that before people kick it out of your house. The real promise of humanoids is one hardware/software loadout that can do many different tasks. But chasing the robustness/reliability asymptote is going to be a lot longer than people think.