The Future of Robotics: Are Humanoid Robots The Right Model?
While humanoid robots are gaining significant investment and attention, we question whether they’re truly the optimal path forward. Discover the potential advantages of specialized robots designed for specific tasks and consider the balance between human-like familiarity and task-specific efficiency. From elderly care to household chores, we examine various applications and challenges in robotics. Join us as we ponder whether society will embrace robots that mimic human form and function, or if we’ll opt for a diverse ecosystem of specialized machines optimized for their unique roles.
Financial analyst firm CBInsights reported in March 2024 that humanoid robot investments had reached record levels at $775M. Significant in that number was Figure’s $675M Series B round from tech titans Amazon, Intel, Nvidia, Microsoft, OpenAI, and Samsung. Figure leverages reinforcement learning, a variety of machine learning, rather than through traditional programming.
Figure’s company’s press release included a quote from Brett Adcock, Founder and CEO of Figure, stating that Figure’s vision “is to bring humanoid robots into commercial operations as soon as possible. This investment, combined with our partnership with OpenAI and Microsoft, ensures that we are well-prepared to bring embodied AI into the world to make a transformative impact on humanity. AI and robotics are the future, and I am grateful to have the support of investors and partners who believe in being at the forefront.”
Many other firms are investing in humanoid robots across a range of applications that include healthcare, manufacturing, retail, logistics, construction, and emergency response.
While I agree with Adcock’s assertion that “AI and robotics are the future,” or at least, a component of the future, I don’t think humanoid robots are the best answer to most industry use cases.
What humans do
As humans, we build our world to accommodate us. We craft living and work environments made for hands and feet, for tall backs and strong shoulders, for binocular vision and ten fingers, for stereo hearing at a narrow range of frequencies, and for eyes that perceive a limited spectrum. We assume we will walk from place to place. Job descriptions, for instance, often suggest that roles require applicants to be capable of lifting 50 pounds.
We can reach above our heads, bend to the ground, and even crawl. We can climb to some degree. We can stir, chop, clap and pull triggers.
And the list goes on.
For my human readers, I need not continue. Each of us knows what our bodies can do, though we each harbor distinctive abilities and constraints.
The question is, should robots do things the way we do them, or should they diverge with designs aimed at specific tasks?
What humanoid robots do
Humanoid robots are designed to fit into human situations. They are designed to navigate our spaces, interact with humans, and perhaps most importantly, use our tools.
All of those very human tasks listed above require humans to work in their current configuration.
Humanoid robots are being designed, for the most part, to be able to fit and work within human constraints. But not exclusively. Robots, for instance, will likely be able to lift objects much heavier than a human could lift, even if the robot is constrained to the same reach as humans.
Perhaps more than the ability for robots to use our tools and fit into our spaces, many of us also have a need for the familiar, for a relationship with the thing that exists in our space and complements our lives. Dogs and cats, goldfish and frogs, all have some familiarity. They have eyes, legs, and faces. They eat and sleep. They exist in our spaces, though they may not be as limited as we are in some ways, such as cats curling up on the top shelf of an open closet or lizards running under and into a couch to hide.
Animals are not humans, but, at least for me, vertebrates offer some level of familiarity. Most people find relating to a dog much more natural than relating to a tarantula or a crab. We may keep many types of living things as best friends, but our relationships require either of us to project a connection to the animal or for a mutual bond to form.
One of the reasons for humanoid robots is the potential for bonds to form. While many science fiction programs show humans treating even the most human-looking robots badly, the makers of humanoid robots recognize that a robot that shares human characteristics may make for a more viable coexistence. Those creating robots for human care imagine bonding, perhaps even empathy, if not love, for robotic companions.
In Japan, experiments are already underway to explore robotic elderly care (see “Inside Japan’s long experiment in automating elder care” at the MIT Review).
Robots will be able to do all of the things we do. They can do some of them better, even if constrained by human body models. But when we explore robots, we need to go beyond human forms. The real robotics revolution may not be in humanoid robots but in robots that can, as evolution does, find ways to serve specialized niches rather than act as general-purpose machines.
What robots can do
While humanoid robots will be able to do many things, they will not be able to do many of them as well as specialized robots.
Take the most common non-humanoid robot, the robotic vacuum, as an example. Robotic vacuums do not look like Rosie from The Jetsons, moving her upright vacuum around, mimicking human behavior. Humans, even upright robots, are not good at vacuuming. Vacuums made for humans are, in fact, not very good at cleaning. Vacuums require people who want to clean.
But even if robots eliminate human procrastination and clean every day, vacuum cleaners made for humanoid bodies are best left to wide, unobstructed areas. Anyone who has ever attempted to vacuum under a couch, dresser or a hutch knows that neither humans nor vacuums are really made for that chore. Furniture must be lifted, or the vacuum suction must be shoved under it. Small areas require attachments, or they just get left unvacuumed. High ceilings are also an issue, as are the tops of cabinets.
Then there are the hoses, attachments, canisters, bags and other components designed to be assembled, disassembled and cleaned by humans.
However the robotic vacuum abandons all traditional vacuum designs. It lies flat on the ground, near the dirt. It uses insect-like brushes swirling toward its sucking maw to bring up the dust and particles from the floor. It can go under most furniture, even couches with skirts. It maps its world to get to every inch of the floor.
Current robotic vacuums, however, aren’t good enough. They can’t get into all corners. They can’t climb up or down stairs. They don’t work on the tops of open cabinets nor do they suck cobwebs from the ceiling. People who own a robotic vacuum still need to own a traditional vacuum or employ someone who does.
If we take the robotic vacuum as a starting point and extend its capabilities based on cleaning surfaces as a premise, then robots could eliminate the lowly vacuum, but not with a generalized robotic vacuum, but with a suite of them.
Homes would need a vacuum to climb walls and suck with their bellies as they roll over dust attached to paint and brick. The top of a cabinet would need a robot nimble enough to avoid plates or boxes, or other decorations—to sweep smallish flat areas and to poke into very tiny areas in between. The same robot might be ideal for cleaning a bookcase filled not only with books but with assorted artifacts and collectibles. Another robot might do nothing other than stairs.
Now, could we imbue the common disked shaped robots with some of those features, of course? But a robot made for cleaning broad open areas most of the time would be suboptimal if it also cleaned stairs, as that feature would only be used a small percentage of the time, leaving other features dormant.
What robots can do is almost anything we imagine, but if we concentrate our investment on humanoid robots, we will suboptimize them. We may create robots we can relate to, but we will miss the opportunity for robots to become tools that do things we cannot, either because of our body designs or the safety of those designs. A humanoid robot driving rivets on a girder would appear slow and awkward compared to a robot that did just that one thing.
Suboptimizng robotics
Scenario planning examines the relationships between social, political, environmental and economic factors, not just technology. That means that the complexity of decisions may lead to suboptimal designs. Just because specialized robots are better at most things doesn’t mean they can do all things. We will invest in the best robot designs. For social reasons, we may accept humanoid robots as the de facto standard.
We can imagine how robots might change how we experience the world. At night, as we sleep, little robots may clean the soffits above our heads. Some may read that more as an invasion than hired-out work. Humans may also be more accepting of robots that exist to some degree within human constraints than those that clearly outperform at specialized tasks.
I hope we choose not to suboptimize robots. That approach will also suboptimze us. We will be curtailing our creativity to work on robots that look ever more like us, rather than looking at the work that can be done and finding ways for robots to do that work safely, regardless of their appearance.
Humans are a general-purpose design. Evolution gave us tools for accomplishing various physical and mental creation tasks. But it did not optimize us for all of the attendants of that creation. How we clean, care for, reconfigure, and populate with large objects, as well as other things we do as humans in our spaces, need not be ours to do, nor are they tasks best left to inventions that look like us.
Robotics should be focused on what needs to be done. George Lucas pointed, perhaps unintentionally, in this direction with R2D2 and C3PO. R2D2 was an astromechanical robot or astromech. It repaired spacecraft in-flight. Lucas and others often stretched its utility for storytelling purposes, but R2D2 had a defined purpose, and within the Star Wars universe, a shape optimized to its work. The same was true for C3PO, a humanoid robot designed as a protocol droid by humans to interact with humans and other species as a representative of humans. I would like to think each species with robots would create a protocol droid that reflected their unique physical characteristics.
C3PO demonstrated many times that he was incapable of fixing just about anything, and even if he were capable, he was not interested. Fixing things was not in his programming.
Our best future imbues robots with the capabilities to perform unique roles within the human-AI ecosystem. There will be roles for humanoid robots, but that body plan will likely be far outnumbered by small robots that live in the nooks and crannies of our homes and lives, autonomously working on our behalf or ready to be called upon demand to lift a heavy desk and put it into a new place.
During a move several years ago, I lifted a television above my head and placed it on a shelf, out of the way of the other accumulating clutter. I heard a whiz and a slurp as my bicep rolled up into my shoulder. I had exceeded my design. My bicep detached from its insertion and bounded up my upper arm, rolling into a ball.
I wish I owned a robot that did little more than move things around. It could have lifted that television to that shelf without incident, leaving my arm without a scar, albeit with me short of a life lesson story.
Combine that kind of robot with AI, and I probably could just have said, “Optimize this room for moving to a new home,” and walked away. Not only would my arm be intact, but I would probably have saved hours worth of rearranging a garage full of things that ended up in a new location in an arrangement no more logical or necessary than the one from which they came.
Sure, a humanoid robot could lift and pack with all the acuity that humans do. Or we could have robots that moved and boxed, complete with tape dispensers and label makers built into their undersides, with arms that could retract and reach and lift.
Should we concentrate on making humanoid robots with the need for general purpose and all the expectations that come with human form? My answer is no. Humanoid robots are not the future of robotics, but that’s just me. We’ll have to see if society wants robots that look like us and, in many ways, work as slowly as we do, or if we want robots optimally designed for the work they do. How would you vote?
AI icon by Siipkan Creative from Noun Project (CC BY 3.0)
For more serious insights on AI, click here.
Leave a Reply