Will robots inherit the earth? Yes, but they will be our children.
I. The rise of robots
Ever since I was a kid, I’ve wanted robots. I'm not alone in wanting this either. Robots are the basis for any utopian vision of the future. And as we're starting to see this future coming together, with AI able to write emails and code and make art, this is the part that's still missing.
Yes it's hard, and yes we've tried in bits and spurts with limited success, but many many things are hard and we do them. So why not this? All we're left with is a lonely roomba getting stuck on the odd carpet or litter box.
When I tried to ask why this isn’t here yet, the major objections were:
The cost is too high
Real world is too hard to navigate
Hardware isn’t ready eg tactile sensors
Battery density is too low
This devolves to a function of two problems - they don’t work very well, and they are not affordable.
Let’s have a look.
They have to work
One of my favourite things I did was to teach a robot how to walk, back in 2005. At that point, since working AI was a pipe dream, you had to think about the mechanics of the joints, angles of bending, weight distribution and centers of mass, inclines of the path ahead, and a whole bunch of other things to be able to create some boundaries within which the robot learns, to train its neural net.
(Parenthetically one part I remember is naming a bunch of variables after characters in the Prince of Persia videogame, just because we were incredibly cool in college.)
And turns out that this is incredibly hard. There is no way you can handle all edge cases when the training is pretty specific, and rooted in biomechanics. To be successful, they have to work relatively autonomously, navigate its surroundings, make autonomous decisions, and be able to actually handle things like a baby blanket or a hot cup of coffee.
To actually do them in real life requires some form of reinforcement learning (I wager) to get the inner equilibrium etc right. I assume this isn't trivial, though I've only done it the hard way. But OpenAI did set a benchmark for robot dexterity, and later abandoned its research because there was no large enough dataset to help power its reinforcement learning.
And the robots of today seem pretty similar. Boston Dynamics have some of the best videos of robots doing cool stuff I’ve seen, but their training method seems awfully similar to creating tightly bounded environments where the robot can learn particular tasks. What can the robots do? Today they only seem to work in highly constrained environments.
To make something that works in the physical world is hard.
Thanks for reading Strange Loop Canon! Subscribe for free!
A big complaint about Boston Dynamics’ beautiful dog Spot is that the battery lasts about 90 minutes. Which is not enough to be able to get almost anything done. Which also means even if you bring Spot home, it’s not going to be all that useful!
This is comparable to the Roomba, also at around an hour and half, at which point it has to stop and recharge. I mean, I can’t blame it, this is already far above the average human.
But it does feel like this isn’t enough. And this means we need a way for Spot to run about the house doing things without having to rest regularly. And that’s having far better batteries. Luckily, Noah seems to think, as do I, that we’re entering an age of the battery. This is essential if we are to dramatically increase this by an order of magnitude.
In mobile phones, for comparison, if we look at the past two decades, we moved dramatically in terms of battery capacity, but increased way more in terms of capability, which made it seem the battery life had fallen. This would be an acceptable compromise, though even with that 90 minutes seem much lower.
But even if we fix the battery problem, do these robots actually work?
The good news is that we’re seeing progress! Robots are becoming more dexterous. The difficulty of real world can be reduced at least a little bit by focusing on smaller use cases than creating a robot Jeeves - start small by picking up and doing laundry! This is also far safer, since the level of damage it can do is far less, and it can move extra slow if need be. And the battery density point is important, though at least inside a household, it’s fine no? You’re never more than a few feet from a socket anyway.
I feel if we had the will there’s surely a way.
They have to be affordable
Cost curves are a fact of manufacturing. The best example is solar power, which has seen reductions in costs that boggles my mind every time I see it. It puts Moore’s Law to shame!
This has come about with a truly gigantic level of investment in producing it, which only got filled because of the demand for it. Without mass manufacturing to meet demand, the costs would never have come down.
Or computers, where Lisa, when it was first debuted to the world, cost $9995, around $30k in today’s money. It didn’t sell very well either, but it started the bending of the curve.
So why hasn’t it happened for robots? My guess as regards an answer takes inspiration from an adjacent world - home appliances.
Home appliances are still insanely expensive (I’m still scarred from a kitchen remodel we did last year) and they are expensive partly because every single one seems to be in a slightly different size to the other, which is a different size to your kitchen, which is a different size to the counter, so the whole thing is like playing Lego with bricks that don’t quite fit together.
This is despite many a mild mannered intervention from the powers that be.
in the United States, the National Kitchen Cabinet Association (NKCA) has established guidelines for the dimensions of kitchen cabinets and appliances. These guidelines are intended to ensure that appliances and cabinets are compatible with one another, and that they can be easily installed and used in a kitchen. Additionally, in Europe, the European Committee for Standardization (CEN) has established a number of standards for household appliances, which include specifications for dimensions and other technical requirements.
This all sounds good, and maybe is true, but from a consumer’s POV, this doesn’t seem to work at all.
This is despite the fact that home appliances market is like $400 Billion and growing about double or triple the rate of GDP growth.
Robotics, by comparison is around a tenth the size, though growing faster. Most of the robotics spend are driven by industrial applications - manufacturing, healthcare, agriculture and logistics - and very little as collaborative robots, or cobots.
Whether it’s linear robots like gantry, or Scara, or delta, or cylindrical, industrial robots have had specialised forms and are complex to install and use.
Hyundai invested $400m for a robotics division. Ford has robots do painting and body construction for its cars, as does Maruti Suzuki. Mitsubishi has robots which the workers can fine-tune with voice commands.
There are grocery warehouses the size of seven football fields which have thousands of robots running on rails to grab, pack and retrieve products, by Ocado, near London.
The “Collaborative Robots” market, which are the robots we are talking about, to run after us in our homes and pick up laundry and do the dishes, that’s a much smaller segment, around $700 million, a large chunk of which is made up of roombas and automated lawn mowers.
There was one study I found of research into what prices we’d need to hit. And they, I have to admit, seems about sensible.
The ratio of the price ranges of the $300-$500 and $500-$700 was fairly high. The noteworthy aspect was that the working moms tended to evaluate the prices to be expensive; the response ratio of the working moms in the price range of $700-$900 and $900-$1,100 were higher than the other targets.
II. What if … we are surrounded by robots pretending to be not?
There is another argument that hey, we’re surrounded by robots, we just don’t call them that. Your coffee maker makes coffee instead of a Barista, your washing machine washes clothes instead of you, etc.
The problem I have with this definition is that while there is automation of some activity, we’ve done so by highly constraining the problem. And as a result, there’s an enormous amount of effort that’s needed to fix, clean, maintain or generally deal with the idiosyncrasies of an automaton.
A robot should, at the very least, empty its own trash like the Roomba does. I mean, that’s just basic.
A robot should have some autonomous capabilities of dealing with an unpredictable environment. A Roomba, in this definition is a not-very-smart robot. An industrial machine that mills complex parts for a car chassis, is a very-smart non-robot. There are many ways to define these things, and plenty in a grey area, but this is my definition. It agrees with intuition, and I’m sticking with it.
More importantly what you’d want is a robot that trundles around the house, picks up and puts away toys my two boys regularly create obstacle courses with, and picks up the laundry and puts them in the washing machine.
This should be doable, no? This is the dream. Without that we’re puttering around in the margins.
Here are a bunch of robots that won the DARPA challenge, flying ones and four legged ones and track bots, operating well in the underground world. Helping to map, navigate and search underground environments, with limited visibility and freedom of movement, and a fragile environment all around.
Tesla is building, apparently, a multi-purpose humanoid robot. I’m not sure why, or even if it’s actually true, but it was announced. More prosaically, Amazon bought iRobot, and Dyson is planning on building household robots. There’s clearly interest. And Goldman Sachs, through some excel magic, think there’s a $6 Billion market for humanoid robots. It seems small to me, but it’s clearly big enough for the largest automaker.
There were also “9 million domestic robots” that shipped last year. Most of these aren’t what we’d think of as robots though. They are automatic vacuums or automated pet feeders or robotic pool cleaners and lawnmowers.
To recap, a robot, to be successful, would have to be able to solve a few hard problems:
Move autonomously, like a Roomba, or one of those Boston Dynamics parkour bots. This is (famously) hard, as the autonomous driving efforts have shown us. Can it be done inside a home, or a kitchen sitting on a countertop? Feels easier, sure, but “feels” isn’t enough to spend $1B in R&D. But the inside of a home, especially is circumscribed, seems a smaller number of crazy variables than handling a 3000 pound hunk of metal hurtling at 70mph.
Be able to manipulate objects, like picking up a teacup or a laundry basket, again like one of those parkour bots. This is hard because of the weird shapes or materials that objects have, as the ability to handle things isn’t developed yet. This is exceptionally hard once the robots aren’t affixed to place. But here too, can’t we try and break the problem down a tad? Solve for clothes only, move on to china, or books. And,
Stop when it hits a problem.
The thing that I think of when faced with this problem is that we’ve solved bigger problems before. When electricity was all the rage we literally had to wire up the entire world. And we did. Same for pipes for water, and again for telephones, and pretty much again for the internet. For appliances we covered the world in televisions and washing machines and refrigerators and dishwashers in record time!
My bet as to why this hasn’t happened is the same reason why home appliances still come in random sizes. Without standardisation, costs are too high.
Ranges will be 30" wide. Slide-ins too, but the depth may vary. DW's will be (close to) 24" wide. Fridges will be 36" wide, depending on size. By that I mean a smaller 18cuft might only be 34", a 25cuft might be 36". Height will vary from mid 60" to mid 70" depending on size.
Cooktops will vary depending on number of burners and who makes it. In general, they come in 'classes' like 4-burner at 30", 5-hob at 36", 6-hob at 42". European versions will be metric equivalents which could be a bit larger (800mm would be about 31.5").
Ovens come in 24" (rare), 27" (not quite as rare) and 30" widths. Heights vary a bit, but you're probably talking around 27" tall.
Which is mostly because most of the appliance sales are for existing homes, rather than for entire new builds, and they require the old measurements to fit into the existing spaces. Economics is hard to fight against.
And also, its just uneconomical to change everything to be standard size when you do this just by yourself.
According to the National Association of Realtors (NAR) Research Group, sellers who upgrade their kitchens recover only 52% of the cost, on average.
Spot’s robot dog costs $75k, and this is the only way to bring it down 100x. It’s the only way anything works. Even in software, have you tried linking up multiple devices to your Google Home or Alexa? It’s a nightmare. Nothing actually speaks well to one another.
Back in the day, from the days of Homebrew Club to even the good old days of early 2000s when I went to college, you could go buy the components that you wanted (the perfect RAM, HDD, cables, motherboard etc) and assemble it yourself. Over time, as we standardised the building of these systems, all through the supply chain may I add, the costs came straight down.
For now, with costs being high, the robot market is still small. The same virtuous cycle that solar and chips saw gets vicious when reversed. And the only way to reverse it is through standardisation to drive efficiency, so that we can benefit from economies of scale.
We’ve made progress. We have made robot arms that can pick up dumbbells, powered by water for some reason. We have people creating their own robotic arms. We have dry-cleaning robots, stair climbing robots, and just look at this robotic kitchen, wouldn’t you want a cool version of this that can do more?
But automobiles only got much cheaper because the assembly lines started using standardised parts and a straightforward assembly technique. This is the as yet unrequited dream behind the creation of prefab buildings. Solar panels are another. The certifications you’d get, like IEC 61215, ensure a consistent standard to meet. With minor variations, the sizes are made the same. Module formats are the same, the power usage and output are the same. Connectors and wiring are standardised, like MC4. Add it all together, you get the 60% reduction in cost over the past decade, with a much higher install base. And who can forget, most famously, the shipping containers that changed the face of modern commerce.
We have dexterity. We have reusable parts. We have pretty flexible robotic arms, especially when stationed on a counter. We don’t have a decent assembly line though, it feels like the old computer days of building a PC yourself. But we have high degrees of specialisation, with purpose-built machines all through our homes. And we have the ability to translate spoken word into code.
So this is my admittedly dilettantish suggestion. I am likely to be wrong, feel free to let me know. But meanwhile let’s standardise an assembly line, and go build them. Start small. A robotic arm in the kitchen, or like something that trundles along behind us and cleans up. No more dangling wires and messily soldered boards. Lux will surely fund it. And I’ll buy it, as would I think a large number of my friends. And then let’s go see if we can’t ask literally every single household if they don’t want another convenience revolution, the biggest since the washing machine.
I think they’d say yes!
Thanks for reading Strange Loop Canon! Subscribe for free to receive new posts and support my work.
It appears no-one is seriously working on sensors, particularly at scale. Your hands have thousands of temperature, texture, and pressure sensors, and synthesizing information from the data these provide is crucial for the ability to manipulate the variety of objects found in unstructured or semistructured environments. (If you've experienced numbness from cold or carpal tunnel, you will know this.)
The easy stuff, vision, has been done--possibly, well enough. Time to move on to the real challenges.
We are indeed surrounded by specialised "robots", but the true paradigm shift is if/when we engineer ones with high mobility *and* ability to generalise to new tasks. I think the latter trait will take more time to get right that the former, but eventually we would be able to free up people's time just as other household items were the unsung heroes of productivity increase in past decades. Alphabet seems to be on this pursuit with Everyday Robots, one of their X spinoffs. And it looks like the transformer paradigm that has enabled systems like ChatGPT might be the key to help robots significantly improve their ability to learn new tasks.