Hacker News Reader

Comment by GlenTheMachine | Hacker News Reader

Full dosclosure: I am a roboticist. I've been in the field professionally for about thirty years.

I'm really skeptical that there is anything “after the roomba”. Or, to be more specific, I'm skeptical that technology used in the home will look anything like a robot, a self-contained intelligent machine that moves around and cleans/folds laundry/does dishes.

Ther are a couple of reasons for this. This first is that these problems are extremely hard to solve. And this is not true just in terms of algorithms, but also in terms of hardware. A humanoid robot is as complex as, if not more complex than, an automobile. Currently, humanoid robots go for $500,000 a pop. Even after economies of scale it's hard for me to imagine that getting below $50,000 — much less $5,000, which is probably the price range that makes them viable as anything other than a novelty for wealthy people.

Which brings up the second point. One thing I have learned the hard way in robotics is that you have to understand precisely what your competition is and how much money that competition requires to solve the problem. In the case of an in-home robot, for most of us the competition is our own labor or that of our significant others/kids/housemates. But we are not the early adopters of this technology. Wealthy people are. For wealthy people, the competition is either a live-in assistant or a contract cleaning service. Contract cleaning services pay their workers approximately minimum wage. Minimum wage is a very difficult price target for a robot to meet.

And the third issue is that to first approximation these problems are already solved. And not by what we would consider a robot, but by appliances. My grandmother, for most of her life, kept house without elecricity, and hence without a dishwasher, vacuum, or dryer. These appliances dramatically reduced the effort it takes to keep house, without being robots. I strongly suspect that any cost-efficient solution to folding clothes or taking dirty dishes from the sink and turning them into clean dishes in a cupboard will resemble an appliance. If you want your dishes cleaned robotically, then you don't put a robot in the kitchen. You turn the kitchen into a robot.

View on HN

> you don't put a robot in the kitchen. You turn the kitchen into a robot.

That was my thinking too after reading the article. Why jump through all those hoops, trying to get a robot mimicking us getting around the house and instead just build smarter, more capable appliances.

E.g.:

- having cupboards that double as dishwashers (the robot does the sorting and drying...)

- having wardrobes that double as washing machines (the robot does the sorting and drying...)

- having trash bins that do the seperation, maybe even compression

But in any of those cases, the "robot" could be a grappler on a track that never leaves the casing of the appliance

This is exactly why robots are so expensive. Let's take a simpler example. Let's say you want sun tracking on a solar power installation. You could of course use mechanical linkages, or individual motors, or a combination. This is what's done now, and it's actually a decent part of the cost (so much it's often not done, or people do ridiculous things, like building a 10x10 meter solar panel so you only have to have one huge tracker).

Think about a robot going around that can aim a single solar panel to the sun in 30 seconds. That robot can do 120 aims per hour, and can keep ~400 solar panels at 96%+ efficiency by going around constantly. A collection of such robots provides redundancy (like humans do, if you've got more than one).

Such a solar tracker costs ~$200 at scale per solar panel. Mechanical solutions are cheaper up front, but you'll be fixing them weekly. So a robot that can do this can sell for ~$50000 and be profitable from day 1 (because it replaces capital expenditure). There'll be way less to fix and you can probably just take the robot back to the factory while providing a replacement onsite, which would be a fantasy with the motor based solution. Or you could have N+2 or higher redundancy for large sites and come by for robot replacements once a year providing constant coverage essentially free.

The same with your kitchen. Having a general capable robot can use a lot of equipment to accomplish a great many tasks. Such a robot even means that the difference between a cheap and an expensive, say, dishwasher, disappears to an extent. The most expensive dishwasher won't match the comfort of a cheap one + a robot, and the same is true for all other appliances.

Given that how much is it allowed to cost ? 5000$ seems like a lower bound. I'd be willing to pay that, which usually means others are willing to pay 100x that, even if that seems a bit much (in the West though, perhaps not as much).

>Mechanical solutions are cheaper up front, but you'll be fixing them weekly. So a robot that can do this can sell for ~$50000 and be profitable from day 1

Aren't you ignoring potential reliability issues of a robot though? A lot more moving parts in a robot than a simple motor on a pivot point.

If you truly had a humanoid robot that could do the dishes and the laundry and cook and clean the house, you would have no trouble at all selling it for $50,000. It would sell like hotcakes at that price. People would use financing to buy it just like cars.

A contract cleaning service is cheap, but they aren't in your home helping you 24/7. This hypothetical $50,000 robot would be much more useful and not that much more expensive if amortized over, say, 10 years with regular maintenance.

Also, at that price, cleaning services, all sorts of industries, would buy them to replace human labour.

At 50k/year that's 2-3x minimum wage depending on where you live. That's 3 employees for a year. And in that wage bracket there are no repair costs associated with the humans. Nor maintenance.

Also who gets the robot to the house? And what cleaning company or what ever has the Capitol for a $50k purchase. The competition has their own Transit and start up costs are cleaning supplies.

The parent and grandparent comments to mine appear to be talking about $50,000 as the original purchase price. That’s the way I read it.

If it were $50,000 per year then I agree with you.

Further to this, Australia, where I live, has a high minimum wage at a little over $20/hr casual, and a little under as a fall tint employee with 20 days paid leave and 10 days sick leave per year.

There’s no way I can afford a full time, 8hrs / day, personal assistant.

But you don't. Cleaning dishes, floor and and clothes is not a full time job: I(ve have done it for decades, but now have people to clean my house, cook and wash my clotheres.

So your robot will sit idle most of the day, unless you have a much bigger house or a big family.

And remember that for the next 20 years, the robots won't be able to do many things due to motor skills/autonomy/analysis capabilities. So they may, by as for now can't replace a full time human.

Android are still scifi.

> So your robot will sit idle most of the day

I think you have a "neurotic" cleaner robot working full time to make sure your house, dishes, and clothes are perfectly finished. Part of the package is a cheap third-world worker for an hour to remotely waldo the robot for the difficult parts, and do the final checks.

The robot is bought using finance, like a car. Or bought on a service plan that gives you "free" upgrades every few years.

It then becomes a social norm to have a robot "My, how unclean your home is, whyever don't you get an Apple Dysony like I have".

(I'm unsure what the marketing would be in countries where you don't visit friend's houses...)

Yep, this is what I’m talking about. Well said too :)

VOIP still doesn't work properly in the Valley one time out of two, and you want somebody from the other side of the world remotly handling breakable things in the house ?

The first robot killing the cat by dropping a knife because of a lag is going to be fun.

Sure, but then...

You know, there hasn’t been a technological advancement we’ve managed to refrain from using. So it’s either it’s just a matter of time, or it’s not possible due to some limitation(a).

Let’s assuming for a minute that this does eventuate, and some robot does manage to kill the cat: we’ll get over it. We haven’t regulated the car out of existence yet, and they have a tendency to kill quite the many cats.*

*all analogies breakdown under scrutiny.

Of course. I'm not saying it can't happen. I'm saying it's still scify. We have nowhere close the technology to make it happen yet.

I'm convinced AI and androids will come around. I just don't think it will happen in this decade.

I’m still not completely sold on the certainty of either. Time will tell.

Jeepers - as though the Valley is the centre of the universe?

If latency turns out to be a serious issue, then remote control robots would be installed where latency is reliable, or where home owners can get low latency connections.

Most likely latency will improve for the valley over the next few years.

Certainly VoiP is reliable in many other countries. We use a cheap provider at work (virtually no issues), and I used it at home in 2011 and it was rock solid. And NZ is definitely on the edge of the main networks!

Edits:

PS: second anecdote: I'm on an island in the Mediterranean at the moment, and VoiP seems very usable.

PPS: what do the gamers in your neighborhood do to resolve ping times? They usually are extremely interested in diagnosing the underlying causes of latency and finding solutions (although you obviously need to find a more engineer type gamer than a script kiddie soothsayer type)

We just switched to all VoIP desk phones at work, over our NBN connections. Same at home. Works flawlessly.

Both providers, Telstra at work, and Internode at home, give VoIP traffic the highest priority QoS, as far as I’m aware. It’s been a while since I worked Internode.

I understand that but your assuming you never replace the robot, no downtime, no amortization, no maintenance costs. Which is not reasonable either.

At 3 years you are likely break even with humans with no maintenance costs. At 5 you're likely actually b/e, and may need to do major matinence or an upgrade .

24/7/5 is a lot of wear and tear.

Even if it were 50k a year, that's under $6 an hour, which beats minimum wage in the U.S.

Good observation!

Even if we assume 90% uptime, it’s going to needs some maintenance, it comes out to a levelised cost of $6.34 an hour, plus parts and labour for the maintenance.

This still blows human labour out of the water because robots don’t complain about conditions, and they don’t intentionally run off with trade secrets.

Yet.

They don't get tired, they don't have days off sick, they are repeatable

Automation is good. Assuming the hypothetical robot that can do the required task actually exists, it's probably worth nearer to $100k a year

Not sure if you've worked with machenery before but it takes plenty of time off.

You only get that if you're working round the clock, which none of the examples would do for companies as the robot would need to move around thus having downtime.

Your also assuming 0 maintenance and no charging time

1. You are imagining current technologies at scale, but future technologies are likely to get cheaper and better. Robotics is still a newborn field. Cars have also gotten cheaper and better; the Model T started out around 20k in today's dollars for a 20hp car that went 45mph.

2. You're comparing an in-home robot to a contract cleaner. Robots would be cheaper for the customer on a contract basis, just as a live-in maid costs more than a contract cleaning service.

I do agree that the Jetsons-esque conception of a humanoid service robot is not how things will likely shake out as domestic robotics matures; IMO the humanoid is a simplistic image that demonstrates a lack of imagination.

Sure. Forever is a long time. But don't confuse hardware with microprocessors. There aren't many fields, maybe not any, whee you see exponential growth in capability and exponential improvement in price sustained over decades.

Cars may be a good model here. But I would argue that current generation robots aren't Model T's. 1960'-era industrial robots were the Model T. Current generation KUKA's are 1950's Cadillacs. Cars of today aren't exponentially better than a 1955 Caddy. The period of rapid growth in robotics (on the hardware side) started a while ago, and it's probably approximately linear.

Robonaut is the Lunar Rover. It's awesome, but it isn't at all clear whether any of that tech addresses a commercial market.

There aren't many fields, maybe not any, whee you see exponential growth in capability and exponential improvement in price sustained over decades.

Obviously, the field where exponential growth is/was sustained over a significant period is chips/microprocessors. Miniaturization is well-know as the key to this.

Information processing so-far is all that has benefited here. But existence of biological organisms, constructed from cells existing on a tiny scale, shows that miniaturization could be applied to other functionalities. There's no reason to think we're near that now but the situation does invite us to imagine it in theory.

I recall seeing something about agriculture productivity improving at 1% per year over many many decades - maybe that wasn't compounding though.

The first search result I found had this: "we track four key global crops—maize, rice, wheat, and soybean—that currently produce nearly two-thirds of global agricultural calories. We find that yields in these top four crops are increasing at 1.6%, 1.0%, 0.9%, and 1.3% per year, non-compounding rates"*

Another example is fuel efficiency of cars? Although that has hard limits.

Sure, nothing like electronics, but over decades slow compounding of small amounts does matter to us all.

* https://journals.plos.org/plosone/article?id=10.1371/journal...

A lot of Robonaut's actuator technology ended up in Baxter. Now Rethink's dead, but it did start the whole cobots sector. In many respects the cobots from UR are closer to the model T then 60s era robotics. 60s era robotics is more like the tracked steam shovel.

>A lot of Robonaut's actuator technology ended up in Baxter.

Well, not exactly. Both Robonaut and Baxter used series-elastic actuators. But Robonaut didn't invent them. This is kind of like saying that a lot of the tech from the lunar rover ended up in Tesla's. Sure, they both used electric motors and batteries. But Tesla's existence is in no way dependent on the lunar rover.

Sure, my statement was based on rumors that I heard about Rethink taking a look at Robonaut with the covers off and the actuators looking eerily similar. But yes, Series elastic actuators weren't invented at NASA.

> Cars of today aren't exponentially better than a 1955 Caddy.

You're talking about cars that cost around $70k in today's dollars with 6.0L V8 carburetted pushrod engines that go 0-60 in 15 seconds, no standard AC, 4 speed transmissions....they looked cool, but you can buy a $30k Infiniti nowadays that is lighter, has just 60% the engine displacement, has fuel injection and variable valve timing/lift, goes 0-60 in 5 seconds, has 4 wheel disc brakes, a 7 speed transmission...the list goes on. Oh, and standard A/C.

I'd say that getting 3x the acceleration at half the price, with multiple times more reliability (try driving a 1955 Cadillac 300k miles without rebuilding the engine or heads), traction control, airbags, intelligent cruise control, and all the modern comforts is indicative of rapid growth. And it keeps going! Now we are seeing electric production cars and autonomous street legal vehicles.

Exponential? Maybe not. But aside from computers, it's hard to think of technological developments as exponential. They tend to be intermittent, with advances coming in fits and starts.

And both will get you to work at exactly the same speed. They even get approximately the same gas mileage.

All of the other stuff - better acceleration, air bags, automatic transmissions, A/C - those are refinements on a capability that fundamentally didn't change. They both drive on roads at around 60 mph at approximately 25 mpg.

A '55 Caddy Eldorado will get 10.5mpg highway according to these folks: http://www.automobile-catalog.com/car/1955/326075/cadillac_e...

They're cool cars but today's $70k luxury cars are faster, stronger, more reliable, and functionally better. The old Caddies do look damn good, but I'll take a DOHC versus an OHV any day.

My point is that the technology has clearly progressed. To your point, maybe not exponentially, true. But there has definitely been a whole heap of engineering work over 60+ years that has resulted in unequivocally better, more reliable, markedly different vehicles.

1. That car massively outperformed its competition, the horse. It didn’t buck you off or startle, didn’t get sick and die, didn’t shit all over the roads. It was able to go faster, longer, and more comfortably than a horse. That sounds like the opposite of current robots, which tend to underperform their current competition, as the OP was trying to explain.

2. He also compared them to just people doing their housework. He also pointed out that this would be a toy for the wealthy, and we have to consider what they want. A friend of mine is half Filipino and very wealthy, and her relatives all have live-in maids. The maids cook, clean, do laundry, basic maintenance, and can even drive kids to school. They’re not particularly expensive either (compared to a robot) and you’d need quite a few different robots to accomplish all of those tasks.

Yes, robotics will improve, but I’m not seeing the horizon past which a single affordable robot can do everything from laundry to dusting for cobwebs. It goes back to the original point of the OP, which is that for people who could afford a robot, these are already solved problems and human labor is relatively cheap compared to half a million dollars of machine.

While I certainly tend more on your side of this discussion, I wonder to what degree your first point demonstrates a modern bias rather than an objective comparison.

Cars could only travel on roads, you couldn't go wherever you liked as on a horse. Narrow passages you could previously have ridden down were now inaccessible.

You need to put fuel in your car, which was probably even harder then than finding an electric car charger at the start of their existence. Providing food and drink for a horse was comparatively trivial.

Horses could easily be tied up outside of whatever store or destination you were visiting. What were the social costs of parking a great big machine outside of your local? To say nothing of parking as a town- or city-wide logistical problem.

Horses can in theory, go anywhere, but in practice it’s not advisable, at least at more than a trot. If you’re going at a canter, or god help you a gallop and the horse’s foot finds an unseen and unexpected hole, stone, etc... you’re going to be in serious trouble and so is the horse. Generally you only ride horses at high speeds on well worn and known trails, or roads. So yeah, a horse is better off-road than a Model T, but I’d rather walk unless I know the trail. Trotting for long periods is an athletic thing if you’re riding English, and torment on the groin otherwise. All in all, riding a horse for long periods is not easy unless you’re very used to it, and physically fit. Remember that at the time most horses weren’t being ridden, they were pulling carts or carriages, and they’re not exactly off-roaders either.

Fuel was definitely an issue, but you could just carry tins of fuel and the car wouldn’t mind carrying them. Carrying food and water for a horse is no joke, and horses also need to rest. You also don’t have to take the saddle and bridle off a car, wash the car and brush it down. Horses really are a lot of work, especially if you’re working them hard. Remember that work also comes at the end of a long day in the saddle, which is hard work for you, especially if you’re beating a trail at a trot.

Parking... I have no idea about the logistics of parking at that time, so I can’t say.

A horse produces ~20 pounds of manure and a liter of piss per day. It was not possible to keep the streets clean. Removing that waste was itself a huge logistical problem.

> At the turn of the 20th century, the most malodorous environmental challenge facing the world’s big cities was not slums, sewage or soot; it was horse dung. In London in 1900, an estimated 300,000 horses pulled cabs and omnibuses, as well as carts, drays and haywains, leaving a swamp of manure in their wake. The citizens of New York, which was home to 100,000 horses, suffered the same blight; they had to navigate rivers of muck when it rained, and fly-infested dungheaps when the sun shone. At the first international urban-planning conference, held in New York in 1898, manure was at the top of the agenda. No remedies could be found, and the disappointed delegates returned home a week early. (https://www.economist.com/special-report/2016/11/26/the-futu...)

See also The Horse & the Urban Environment: https://enviroliteracy.org/environment-society/transportatio...

To be sure, I'm not saying cars weren't an improvement over horses (history proves they obviously were), I'm just saying that after a century of car culture we're probably somewhat inured to and/or have engineered around the negatives of car ownership, so they've become invisible to us. Now it just seems like a near-perfect drop-in replacement, which I guarantee it wasn't at the time.

That said, I'm unconvinced that smell and mess are as compelling a selling-point as all that. Societally they're very important, but not obviously more than non-fuel-burning vehicles today. In practice, when it comes to making a purchase, there are often much more pressing concerns than an environmental issue over which you individually have little influence.

1) False

> That car massively outperformed its competition

Nope. Initially cars couldn't even match horses in speed. Although speed did rapidly become the reason to get a car, but initially this wasn't the case.

> It didn’t buck you off or startle, didn’t get sick and die, didn’t shit all over the roads

Yes it did. Early cars ... there were no more than a few dozen in a city, yet it was a daily occurrence that one blew up. Early gasoline engines ...

Second: the brakes.

Third: incomplete burning of fuel. It was constantly leaking oil/gas everywhere.

And all that is assuming you got them started. Even movies from 1920/1930 are full of scenes were cars don't start. That was after decades of fixing issues. Imagine what the first ones were like.

> It was able to go faster, longer, and more comfortably than a horse

Nope, nope, and nope.

Carriage suspensions were far superior to car suspensions for several decades. Cars were big bikes with an engine. More comfortable than a horse on bad dirt roads ? I doubt it very much.

Here's a story on the first long distance car driver (a woman !). Read it, because you'll realize just how bad it must have been:

https://media.daimler.com/marsMediaSite/en/instance/ko/Augus...

That was a rigid tricicle. No tires. Oh, and if you don't know how to make brakes out of old shoes, no brakes, or at least not for long.

We’re talking about the Model T, not “early cars” in general as you’re trying to shift this to. Model T’s were not routinely blowing up, and were quite successful. Let’s leave the goalposts alone.

You're the one who drew the comparison between the Model T and horses, ignoring decades of automobiles in between. That struck me as you moving the goalposts.

Not to mention that even the Model T came out with handcrank-start, which involved manually retarding the spark timing so you don't spark before TDC and break your hand when the crank kicks back. Ford didnt offer electric start until after the first decade of production. And that's in an era where well over a hundred different automobile manufacturers were already in existence, the automobile had already circumnavigated the globe, and gasoline powered taxis were already hauling fares in Paris, London, and NYC.

Have you already forgotten your own post that started this?

https://news.ycombinator.com/item?id=18297710

I didn’t set the comparison, and decades of cars before the Model T (released in 1908, only about a decade after the first practical automobile available in the US was released by Benz for a fortune) wwre hardlu displacing the horse. The mass transition from horses to automobiles began in earnest with the Model T, and the post you were replying to was pointing out the need to define your competition. The major competition for the Model T was a horse. Cars before it were finicky, incredibly expensive, unreliable, and out of the hands of the average person. The Model T changed all of that. Given that we were talking about how current humanoid robots could only be the toy of very wealthy people, it seems a fair comparison. The problem for you is that it supports the opposite point to the one you tried to make.

My point was that robots will improve with time, just as cars have. I still stand by that.

You took my point to mean that robots must today be immediately better than maids, just as the Model T was immediately better than horses. I think this is flawed because robots today are not at the same stage of development vs maids as the Model T was vs horses.

Your comparison of the Model T to horses is flawed. It wasn't the first automobile, but rather the first automobile produced on a moving assembly line. In other words, Ford had found a method to make cars, previously expensive and unreliable, into a viable mass market product. It was an improvement on prior automotive manufacturing technology.

What you're missing is that the Model T wasn't competing with horses, considering that production automobiles had been around for decades. It was competing with more expensive, heavier, less user friendly automobiles. The Buick Model 10 cost more, weighed more, and was mechanically inferior with its exposed clutch and user-oiled valve train. It, too, never got sick and puked on the side of the road. The market wasn't "Horse vs Model T", just like it's not "broom vs Roomba" today.

I don't get why it matters that your friend is Filipino or that her relatives have maids. We all know that some people have live-in maids. The question is, at what price point can comparably utile cleaning robots come to market? There will come a time when it costs less to have a robot (and hire an uber to drive the kids) than to have a person doing the same job.

When the technology brings that price point to a competitive level, the business opportunity will be obvious and there will be more investment. And that investment will bring the price down even further due to broader scale and more research money. Just as has happened with cars.

I think anyone who says robot design and manufacture are not going to encounter the quantum leaps that have graced every other technology known to man, is suffering from a lack of imagination and an inability to examine history. Imagine a group of cavemen arguing that stone tools are cheaper and better and more accessible than bronze because metals are too complicated to dig up and refine, fire is hard to deal with, and as a result the accessible bronze tools of the era are defective and poorly forged. They have a point but they are wrong in the long term. Technology advances and becomes cheaper as people find clever new ways to tackle old problems.

Meanwhile, there are always a bunch of uncreative naysayers standing on the sidelines. Obviously, people who say "this idea will never, ever be viable" are not going to be the ones who innovate and make money. For every Thomas Edison, there are a thousand less productive engineers who get frustrated and say "let's just stick to candles and gas lamps."

> human labor is relatively cheap compared to half a million dollars of machine.

Ask a laid off factory worker what he thinks about this.

Cars have also gotten cheaper and better;

Not by much. Sure a modern car is faster, more efficient and more reliable than a Model T, but at the end of the day it still does exactly the same thing in exactly the same way. Also $20k is roughly what you pay for a new entry level Ford today so it's not like prices have come down dramatically.

A today's car that could be compared to the Model T can be bought for less than $1000 in China. There is no demand for these in the USA but it exists.

One "cleaning" task that I think is often overlooked is basic organization of small clutter and such. It takes up tons of spaces, is often poorly organized, and finding things you lost is a huge headache.

I'd love to see some sort of out of the way storage system with an in-home Kiva style robot that lets you select what you want in an app, and it goes and fetches it. When you want to put something away, the robot shows up, collects your item, and stores it.

To be truly useful, you'd probably need a somewhat decent sized storage area, and the robot may need to be decent sized depending on the weight and size of objects to be handled.

I suffer the same issue. I think I'm just gonna buy cubbies soon.

Clear Hanging shoe racks have helped a ton as well as large Ziploc type bags.

For me the robot vacuum was just a fun way to vacuum. I wasn't making any grand calculation of competition or value added, I just thought it'd be fun to have a robot. Also, I'm skeptical of humanoid robots because companies can just pay actual humans dirt wages and get away with it in most countries.

> her are a couple of reasons for this. This first is that these problems are extremely hard to solve. And this is not true just in terms of algorithms, but also in terms of hardware. A humanoid robot is as complex as, if not more complex than, an automobile. Currently, humanoid robots go for $500,000 a pop. Even after economies of scale it's hard for me to imagine that getting below $50,000 — much less $5,000, which is probably the price range that makes them viable as anything other than a novelty for wealthy people.

I'm curious why you think that is the case? What is so expensive about humanoid robots? I don't disagree with you, I have no knowledge on the matter, i'm just curious as it doesn't comport with my intuition.

Humanoids, in particular, have a lot of moving parts. The human arm has 7 degrees of freedom. The hand has something like 21. Legs, 7. Feet, somewhere between 3 and six. That's a lot of motors, gears, motor controller boards, joint angle sensors, force-torque sensors, etc. Humans have ubiquitous sensing. Tactile sensing of up to 1000 sensors per square inch of skin. Force sensing at every joint. High bandwidth, highly efficient actuators (muscles).

Humanoids are just extremely complicated systems with a lot of moving parts that require very high reliability, along with a metric ton of sensors and the wiring harness needed to support all of it.

For reference, a standard brushless DC motor that is reliable enough, strong enough, and precise enough for use in a robot will retail for around $1000. Just for the motor. And you need at least 30 of them, probably more.

Not to mention that humans run on food, produce little waste heat, are portable, and regenerate many tissues as needed for decades until we enter EOL.

its 100% in R&D. Real commodity hardware is sold at ~3x raw material cost, with R&D amortized over millions of units and very long period. Every time you see something made in china and go "how the hell can this be so cheap" is seeing real commodity price, with no R&D and middleman margins. People will argue about quality, by writing replies on keyboards that cost $5 to make in China, and clicking send with $0.2 BoYue microswitch in their $70(retail) mouse.

You could in theory make 500K robots at 5K if you give up any notion of recouping development cost by tackling ridiculous margins, and instead concentrate on moving millions of units.

Why aren't cars $5k then? Heck most basic motorcycles are $7k plus and they're well understood and very simple.

Cherry QQ (chevy spark copy) is ~$5K in china, being a ripoff removed R&D cost.

Great you literally cherry picked the lowest possible example for your argument. So after 100 years of development and ip theft yes the least featured model might be at that price level. I concede the argument. I look forward to getting my cheap robot in 100 years on the libertarian colony of vega

Cheapest "cars" in China start under $1K, Cherry QQ is actually decent in comparison. Want different example? How about micro controllers? https://www.youtube.com/watch?v=Rixo78hv_lw 3 cent in Asia, cheapest US based manufacturer offerings start at 30 cent.

As a data point. The original Valkyrie cost ~12.5 million. Thats hardware and software dev over about 3 years. I'm not positive about the subsequent copies, but I've heard they're in million to 2 million.

Not the OP but in my experience with lab robotics, locomotion and sensing are the primary problems, as broad as that is. Compared to electrical systems, the human muscular and nervous system are extremely sensitive, low mass, and low volume. The dexterity needed to do everything from assemble electronics to fold clothing while sensing the subtleties of cotton versus synthetic threads is simply not possible with electronics while fitting in a reasonable size.

For example, just the motors, rotary encoders, and joints of a 6 degree of freedom robotic arm capable of lifting 20 kg (with a specialized tool, no hands or jointed fingers!) are 2-3 times the size of the average human arm without any of the sensing capabilities or dexterity. Beating a single human hand in pipetting at speed, for simple vertical movements, takes up even more space with the xyz ballscrews.

Obviously biological systems are much more limited in repeatability, endurance, compliance and so on, but those are the trade offs we get with technology.

Maybe on day we'll get artificial muscles and skin but that is all a long ways off and it might only ever be available to super rich.

Hmm, this makes me think there will be first sex bots that will get some extra new useful function for household in each iteration... VHS story all over again, pushing for economies of scale.

I mean, pornhub have created "The Twerkin Butt", a crude sexbot paired with VR glasses.

> I'm really skeptical that there is anything “after the roomba”. Or, to be more specific, I'm skeptical that technology used in the home will look anything like a robot, a self-contained intelligent machine that moves around and cleans/folds laundry/does dishes.

I think your right there. Personally I don't even have room for a humanoid robot, it'd be getting in my way constantly.

What I would like to see is something more along the lines of a series of those programmable industrial robotic arms extending from my ceiling. That's something that could fetch me a beer from the fridge, take my empty plate to the sink and move my dirty clothes to the laundry, all without getting in my way. That would be a good base to start expanding functionality on.

These tasks are not real problems in need of solutions. Fetching a beer from the fridge, taking an empty plate to the sink, and throwing dirty clothes in a hamper take so little time that there is no reason to automate them.

Those are mostly examples, a machine capable of doing that would be capable of much more. If it can move clothes around it can probably work the washing machine, move the washing to the drier and hang up the dry clothes. If it can fetch a beer it can put the shopping away and maybe give a stir to whatever I'm simmering on the stove top. It could possibly vacuum better than a roomba.

Also for some things like the clothes hamper example, it's not a matter of saving much time but actually doing it where a lazy human (me) would just leave it.

If a robot were to do the entire household, I think people would be willing to pay north of 2k per year for that. If you spend an hour a day doing laundry, cleaning dishes, emptying the washing machine, that's 365*8.5= 3k

I don't think humanoid is a viable form factor. A mobile robotic arm is easier to build and has similar benefits.

It has the benefit, that we already build plenty of them and they have proven their usability in industrial settings. Now we"just" need the brains to run one of these in the kitchen

Robots in industrial settings don't need anything like the capability (sensing, dexterity, etc.) that a useful household robot would need. And they need to be far more safe than an industrial robot, without the safety caging that industry relies on. The problem is at least an order of magnitude more difficult. It's harder than autonomous driving, both algorithmically and in terms of hardware.

There are already robotic arms without cages e.g. https://www.universal-robots.com/products/ur5-robot/

Yes, we currently don't have the technology, but nailing one or two use cases in a couple of years(e.g. being able to pick up and collect clothes onto a pile, maybe even emptying the dishwasher) may be enough for initial deployment.

Also since you can make the hardware quite safe, you can have faster Software Innovation than in cars

>being able to pick up and collect clothes onto a pile

This is pretty much a textbook example of something that would be both very difficult for a robot and an almost trivial task for a human. A washer/dryer already does about 95% of the work of doing laundry. (Less, admittedly, if you iron but then that's another task. And one I assume there is machinery but just not worthwhile at home scale.)

And, of course, there's always the option of taking your laundry into somewhere to have it done for you. There have been efforts at services that pickup to do this sort of thing but most people aren't willing to pay for it.

>Also since you can make the hardware quite safe, you can have faster Software Innovation than in cars

Probably not so. If these problems get solved, they will get solved by an intricate dance between hardware, sensing, and software. This is not like a car, where we mostly know what the hardware needs to look like and it's just a matter of nailing the algorithms. For humanoid robot we may need - will probably need - pervasive tactile sensing, and possibly smell, and certainly visible light sensing but maybe also infrared, or lidar, or sonar, or who knows? We're still arguing about inner position loop versus inner torque loop versus series-elastic-based compliance control schemes, for heaven's sake. And we don't know what the right kinematics for the hands will be, and on and on.

One thing I am absolutely certain of is that no currently existing humanoid robot could be placed in a home and successfully clean it, no matter what software it has.

You may not have to pay for it all yourself. Imagine your neighbors or friends get together to buy or rent the robot. The robot is privileged by your home security system to enter if it is by itself or with individuals that have equal or higher privileges. Your home computer knows your schedule and when you're away. When it's convenient or needed your home schedules a visit from the cleaning robot, possibly transported by autonomous vehicles, and it arrives and performs appropriate cleaning.

If the robot is humanoid or not is beside the point. It may be an expensive robot, but if it lasts years, and can service many households, it becomes affordable.

A 500,000 dollar robot, if it lasts a decade, and can service ten or twenty households is between 2,500 and 5,000 a year, 200-400 dollars a month plus maintenance and travel? It's not inconceivable. If it has 100 operational hours a week it could spend 5-10 hours at your house cleaning a week.

I'm not familiar enough with the price of house cleaning services to know if this a good deal or not. But if that first cost falls from 500,000 to 100,000 or 50,000 then surely this starts to be a great deal.

Then you're still competing with Merry Maids. Which pays minimum wage. Sure, there's a point at which the robot works fast enough for long enough with a sufficiently high MTBF that it becomes cheaper than a minimum wage worker. But it's still a very hard price point to hit.

and you can't tell it simple non-standard instructions as efficiently. 'Leave the sheets in the spare bedroom, and please use our non-allergenic cleaning solution on the bottom shelf in the laundry'

I would more for a robot cleaning service than a human one. The rich stopped using servants in America more because of social reasons that cost ones, although that idea seems to be fading away as Silicon Valley tries to get people to pay to have everything done for them. Call it "single serving servants", for everything.

I like the idea of the shared bot, but the gotcha is that my neighbors and I likely all want the cleaning to happen when we're at work, children are school, etc. Spreading the times equitably means everyone having to put up with it whirring and bashing around in the middle of the night once a week...

So the bot does days cleaning residential areas, then nights cleaning stores and offices.

One robot shared among a whole block doesn't seem realistic. That just doesn't happen, not in the US. You'd never get everyone together to agree on it and all pay their fair share up front.

Much more likely would be a robot-for-hire service, like existing maid services, with the robot being driven in on a set schedule. The robot would have the digital key to your IoT lock and would let itself in.

Uber for Roombas

Maybe not a whole block but I could see 2-3 average-size houses splitting costs and times effectively. Outside of an hour or two of dishes, an hour or two of laundry, and a couple hours cleaning, there will likely be a lot of idle time.

Like, my grandmother has a maid come once a week and the maid doesn't even use up all of her 2-hour appointment getting the place squared away.

I think the cost problem can at least partially be solved by taking the "sharing economy" approach. Eg, you don't own the robot, you rent it.

It still has to compete with minimum wage.

Even at 100% utilization, that robot can't be $500,000

It doesn’t have to pay for itself every year, which helps. (Still too much though, maybe...)

You can get a spot mini for ~30k according to Raibert.

I think limiting our analysis to one particular form of robot is probably not a good idea. As we're still trying to figure out which form is the best for whatever that "killer app" is in robotics.

Also one reason to not turn a kitchen into a robot is generality of solution. If you can have 1 robot work in 3 kitchens is cheaper than retrofitting the 3 kitchens.

All of that being said I generally agree with your sentiment, and I think we'll see robots move into places they're needed/bring more additional value than general house chores. For example, restaurants. Where they can be faster, safer with food, but still general enough to put cheese on one burger and not the next without an engineer coming in after installation.

You wouldn’t have to buy a humanoid robot; you could rent it, and given the typical rental factor, the monthly price wouldn’t be far off that off in-home domestic labor. Less when you account for labor law compliance.

I think you are looking it the wrong way. It's very easy to come up with things that are not possible, but not very useful exercise. Technology evolves in the direction of what is possible and useful now. The original Roomba came into being when it was possible for something as dumb as a doormat to navigate the house.

Other applications of the same algorithm are already happening - Husqvarna uses the same random walk to mow the lawn, and there are other robots that will weed the garden. I have a retired engineering mate who armed himself with 3D printer, a CAD package and an eBay account to construct his own automatic lawn mower. It doesn't use a random walk - it uses the 8mm precision of an RTK GPS to go through a planned an arbitrarily complex path. 8mm is good enough for all sorts tasks of course - like hedge trimming, cleaning outside windows and using compressed air to clean gutters.

Inside the house Roomba's mates have already evolved. They now know where they are in the house, which was introduced to allow them to resume where they left off. But it has also meant they can have virtual boundaries, which means the can be told where the border between time tiles and barriste carpet is so some of them are sporting mops now. Cleaning walls, side webs of ceilings and bathrooms can't be too far away. Dusting will follow shortly thereafter.

The point is these things evolve. I don't know what point you say it isn't a Roomba any more, I daresay it won't be obvious - I'm sure you will say the current lawnmowers are just Roomba's with blades. However while something that puts a wet rag on a wall won't be much smarter than today's vacuum cleaners, it won't look like a Roomba.

I think the explosion week come with imaging. I don't mean image recognition - just the recognition of 3D boundaries of objects - things like walls, bench tops, obstacles on the floor, where the windows and mirrors that need to be cleaned are, where cupboard doors that can open and shut are, where dishwashers that contain round flat things like plates that always go into the same place into those cupboards are - this is how things will evolve.

I agree we won't see a Boston dynamics style robot in the house time soon. What we will get will be far dumber, and far more useful.

I thought about it before. I'm just waiting to spend between 5-15k for a robot who cleans the kitchen and does laundry.

After the autonomous car phase will become boring people will spend more in robotic.

Thanks for this write-up, very insightful

What if I told you that these robots were assembled almost entirely by other robots? And that their raw materials were mined by other robots?

My first response would be that prices will always equilibrate such that complicated machines are always expensive relative to income.

Except for microprocessors. But microprocessors are deeply weird, economically.

Nanotechnology must be the answer, then.

The foundries operate at very low margins, They need atleast 5 years to get the money back. I think it was a race to lower costs as there is no significant change in quality of chips between taiwanese foundaries or former german foundaries.

What does that mean though? Are you suggesting that there are other problems that are already done by robots which in consequence means that these concrete tasks will also be done by robots at some point?

How does that have anything to do with the concrete arguments that the parent comment is talking about?

So what? The FANUC robotics factory is a completely automated lights out factory, yet even a half meter arm that can brealy lift more than 10kg goes for 20-30k. Robots are just expensive.

False! Robots are expensive for three reasons 1. Small scale 2. High precision and 3. Patents. By far the largest one of those is scale.

If you can build a car for 15k you can build a humanoid robot for 15k.

I'm happy to take a long bet on this.

Existing industrial robot arms are already built at some scale, and they go fo $50,000 and up. And that's just an arm. No tooling, no sensors.

This may be that the economies of scale aren't big enough, but we're already talking about hundreds of thousands of units per year in the industry. The car industry builds 75 million or so units per year. They get away with it because of turnover. Fleet owners don't keep cars past 20,000 or 30,000 miles. High end car owners replace their cars every couple of years. They pay $30,000 to $70,000 every five years. Because cars are more than transportation, they're a lifestyle choice and in many cases a sign of conspicuous consumption.

I have a hard time thinking of any other consumer item that is both owned by the vast majority of people and is also a sign of wealth, that is purchased in the high tens of millions every year and for which people happily pay more than $20,000.

Cars aren't a good proxy for household robots here, I think.

"Fleet owners don't keep cars past 20,000 or 30,000 miles. High end car owners replace their cars every couple of years. They pay $30,000 to $70,000 every five years."

Cars aren't scrapped after five years. The average car on the road is over 10 years old, and they're not normally scrapped before about 15 years old.

>Cars aren't scrapped after five years.

No, of course not. But fleet owners buy new ones every five years or less. Typically the old cars are sold on the used market. Which means there is always a market for new cars, and a steady availability of used ones.

But the point is: fleet owners turn their inventory over much faster than the cars wear out.

That $20-30k isn't what they cost to make, though. It's what they sell for. Prices in the free market are set by the second lowest supply cost.

At volume (e.g. an order of 1000 exactly same robots) you can get a 2m industrial arm with 200kg of payload, 2m/s speed and 2m/s2 accel for around 10k€. It's close to the material cost, that's why robotics companies venture off into new field such as collaborative robots, medical robots etc. where margins are still higher.

These robots operate with much greater precision and have much higher durability and strength than required for a home robot. Additionally, people won't need a robot in their home 24/7. Just for a few hours per day. The robot can transport itself between homes during the day, splitting the cost. Additionally, a lot of that cost is engineering. If you only made a few million iPhones, they'd be much much more than $1000.

The limiting factor will be how quickly AI that can support such a thing comes along.

A friend's family "solved" the problem of cleaning dishes and putting them in the cupboard by having two diswashers, a magnet that said "clean", and a magnet that said "dirty".

So they just eliminated the cupboard? They must not have a whole lot of dishes!

Looking at the existence of caches in our CPUs and computers (and all CDNs), I conclude that the 80/20 rule apply to dish washing!

(/s)

But really, it might just be that people use 20% of their dish-washable objects 80% of the time. So having 2 dish washers eliminate 80% of the labor of putting them in the cupboard because 80% will be gone in the other machine by the time you really have to fill the cupboards.

Then again, that's lazyness and I suspect the higher humidity of the dishwasher is an health issue.

No, in fact they had two self-cleaning cupboards.

That seems like a fire hazard.

It’s literally full of water.

It was a joke.

Hell we've only ever had at most one dishwasher, and Mrs Funk is always treating the bloody thing like a cupboard.

I love this idea. My wife and I are terrible about unloading the dishwasher and usually end up using the dishes out of it and piling dirty dishes in the sink for a few days. Then when the sink is full, it's more work to unload the few remaining dishes and then load it with dirty dishes.

I'm now seriously contemplating putting this on a my list for when I renovate/build a house.

If I don't like it, it's also pretty easy to rip it out and replace it with a large cabinet, especially if you opt for the "built-in" door cover for the dishwasher.

Hah, I also know someone who did this. To make it work well you need to downsize the number of dishes you have (even without two dishwashers this is worth it) and train everyone to never leave dishes in the sink.

Most major manufacturers offer this now, in a single washer form factor. Look for "drawer dishwasher" or "dual dishwasher" on any site that sells appliances.

This is genius! Perhaps they should split dishwashers in two, with an option to only clean one half at a time.

I can't tell if you are serious. But this has existed for a long time.

I am serious! I know what you mean about top and bottom drawer separation, but it's not the same. You don't want your dirty dishes above the clean ones, plus you still want access to clean dishes without interrupting a cycle. It's a subtle change, but the UX for me personally would improve quite significantly!

The parent said they already exist. Fisher makes DishDrawers: https://www.fisherpaykel.com/us/kitchen/dish-washers/dishdra...

Ah, I had honestly never seen/heard of them, so I thought the parent was referring to washing the top shelf vs the full DW. Now I know - it's something to add to the wish list :D

Brilliant. By being simple.

Reminds me of - and can be combined with - my father's "invention" when he got a bad back of installing the dishwasher at cupboard height.

Perhaps it'd be possible to have a 24/7 washing cycle, with just-in-time drying when anyone reaches into get dishes.

Econo european models for washing/drying clothes take 6+ hours for a reason.

Reply on news.ycombinator.com