We Aren’t Going to See General Purpose Robotics, and Here’s Why

Ryan Hickman
8 min readMar 26, 2019

--

I’ve always had a fascination with Disney’s Fantasia. Of course I wasn’t alive when it came out in 1940, but I’ve watched it countless times and dreamed of a day when I would have the power of Mickey Mouse in The Sorcerer’s Apprentice. Who wouldn’t want to wave their hands in the air and have our world come to life in order to do our work?

Mickey Mouse from Disney’s Fantasia

The story comes from the German poet Johann Wolfgang von Goethe, who proposed this magical method to enchant everyday objects in 1797’s original The Sorcerer’s Apprentice, long before the Microsoft Kinect tried to enable similar gestures for control. Of course, the real drama of the story is how everything goes wrong, which is part of what makes it so enticing to think about.

Illustration from around 1882 by F. Barth

Fast-forward 222 years, and a lot of that magic is real

Most of you reading this already have a supercomputer in your pocket. Your home and office is also likely full of objects that react to simple touch gestures, or maybe your voice. Today’s magic spells sound less like pseudo-Latin phrases, and more like “Alexa, what’s the weather like tomorrow?”, with no “Abracadabra” required.

So why can’t Alexa do my laundry yet?

In magic lore, there’s an assumption that spells and and their power affect the entire “essence” of the object. Physics is usually thrown out the window and mops and brooms can twist and contort, or outright fly in defiance of gravity.

In the real world we’re stuck with atoms, molecular bonds, and the all-to-often boring reality of physics. This is precisely where robotics struggle to find a common interface, an issue without any hope of generalization on the horizon.

The information realm makes flexible adaptation a breeze by comparison. The device you’re reading this on has a general purpose CPU inside, and it retrieved the text of this very sentence from a remote computer, which also has a general purpose CPU in it. The article was written on yet another machine, and the next reader after you might view this on a computer made by yet another manufacturer, using a CPU from an entirely different vendor.

There’s a ton of standards that make that possible, and it’s easy to be deceived into thinking these standards could be the key to robotics. This imaginary dream of general purpose robotics lives on, and many people think that we just need the right universal operating system, or a widely agreed upon standard software framework, to make it all possible. Some think we just need to replicate the impact that the personal computer had on information systems, so we could enable robots to become part of our every day lives.

Sadly, that analogy doesn’t work. Information manipulation and physical manipulation are just not the same thing at all.

It comes down to those pesky physics, and a recognition that human social evolution stumbled onto something fantastic

No other species we’ve encountered has accomplished as much as humans: we successfully broke down the complexities of the universe into tokens of information we call words. They first were expressed through unique sounds, and then later we added written symbols for those sounds. As time went on we created alphabets and dictionaries and then entire encyclopedias of knowledge.

In the natural word, trees don’t offer their age as an integer, yet we found a way to count the rings. The sky doesn’t know it’s blue, but we can describe its color differently than when it’s dark and stormy. The sunrise has no awareness of time, but humans do, and we found ways to measure and plan around it.

We found a way to make very abstract and complex aspects of the world manageable by breaking them into small pieces that our brains could understand and remix into new ideas. We manipulated raw information, mashed it up, record everything, shared knowledge, and leveraged it all to move the world forward.

Enter computers. They started with such a simple premise: store and retrieve alphanumeric characters for documents and perform simple math. It was our language, both in words and numeric symbols, that enabled early computers like Konrad Zuse’s Z1 to emerge using Boolean algebra. This mechanical marvel just needed to think about the world in tiny true/false decisions.

Over 80 years later, that modern supercomputer in your pocket is still processing software as 0’s and 1’s. That evolutionary marvel of human language has allowed us to distill all the world’s information into these 0’s and 1’s and build up our entire digital economy from there.

Yet, despite thousands of years of effort, we can’t manipulate the physical world in the same fluid way.

We just don’t have the magic to move masses at the atomic level

Any fan of robotics is likely a fan of How It’s Made. Few shows have ever demonstrated the challenges of manipulating the physical world as thoroughly as this one. Episode after episode dives into the unique challenges faced by companies across a huge range of sectors. Food processing, farming, furniture, clothing, and automobiles, have all been covered in episodes that blitz over the unique challenges into making the products we use every day.

The information about those varied manufacturing processes could be captured by a digital camera and transmitted to a digital TV. That’s the flexibility of information distilled down into bits. Sadly, the physical world distills down into atoms, and we’re nearly powerless at manipulating those one at a time. Just look at the latest robotic solutions for automation that I captured at the Automation Technology Expo (ATX) West conference.

2min video full of robots captured at ATX West

Without the Star Trek tractor beam’s ability to directly move masses, the robots of today rely on simplistic grippers and suction cups to affect the world.

Information also has another leg up on physics

How do you move a virtual “mountain of information”? One bit at a time.

A wide number of computing solutions don’t have to run at real-time; batch jobs and overnight processing are just fine. The flexibility of speed for information systems allowed computers to start providing value even when they were orders of magnitude slower than the Gigahertz beasts we have today. It just took more time to get the job done.

Now, try slowly moving a large physical mass with an underpowered mechatronic device. It’s not gonna budge.

Physics has that annoying moment of inertia thing, where a minimum amount of torque is needed before a rigid body will even begin to move. For example, the high torque engine in my car is overkill to mow my lawn. A mower engine from Briggs & Stratton doesn’t make anything small enough to drive a screw into drywall. For that we might use something from DEWALT, but even their smallest electric driver is too big to fix my smartphone. Our homes are full of powered gear unique to the job.

When we look at robots we DO have in the home today, iRobot sells an entirely different line of products to vacuum than it does to mop the floor. They’ve been in business for 25 years and can’t even make a general purpose floor cleaner. Ironically, you CAN ask Alexa to clean your floor, but that’s just sending information to iRobot’s cloud stack on top of Amazon’s general purpose compute systems.

Getting back to why Alexa can’t do your laundry, it’s because the mechatronic system to wash, dry, and fold clothes, hasn’t been invented yet. You can be sure that when it does, it will be entirely focused on laundry like a FolidMate, so don’t expect it to also wash your dishes.

Bespoke robots win the day

My favorite portrayal of our robot future again comes back to Disney. In the 68 years following Mickey’s use of magic in Fantasia, we start to see an entirely different image of how we might control our world in 2008’s WALL-E.

Many robots on The Axiom from Disney’s Wall-E

The creative minds at Pixar imagined the world of 2805 as being full of unique robots, highly tuned to perform their jobs. There’s a still a bit of magic as they seem to float in defiance of gravity, but their physical interactions in the world are performed by specialized tools.

WALL-E also showed us ways that things can go wrong, of course, which is what makes it so much fun to watch, as the premise of The Sorcerer’s Apprentice remains: humans want power over our world and expect it to bend to our commands. This age old desire gave us the word robot in the first place.

Humans on The Axiom taking advantage of robots in Disney’s Wall-E

So why not make humanoids that use our tools?

I don’t know about your home, but my garage can’t fit all of the equipment that a plumber, electrician, landscaper, or other service provider would have in their trucks. Any home robot limited to using a reasonably sized cabinet of tools kept onsite would result in a lot of compromises in performance and quality. Just like when I try to fix things myself instead of calling in an expert.

Owning a humanoid robot in the home and giving it access to an entire Lowe’s or Home Depot via delivery robots isn’t the answer either. I’d rather order a Carpenterbot, Electricianbot, or Plumberbot to arrive on demand and be the perfect size and shape for the job to get done quickly and at the highest quality.

I’ll share more thoughts on humanoid robots another day, but let me just leave you with this. It’s the PR-T beaticianbot, also from Disney-Pixar’s WALL-E. We can assume it has some general purpose compute inside, sensing who you are and understanding what you want done, but then gets to work with a unique array of purpose-built end effectors.

The PR-T beautician robot in Disney’s Wall-E

Until we find a magical way to manipulate matter in the same generalizable way that we do for information, PR-T is about as close as we can expect to get for general purpose robots. Hopefully we don’t have to wait until 2805.

--

--

Ryan Hickman

Robotics startup founder; Ex-Googler; Husband and father of two; loves the future where hardware comes to life thanks to AI. https://twitter.com/ryanmhickman