Hacker News Reader

Comment by freehunter | Hacker News Reader

Of course a robot could do those things. Those are very easy things to do. The problem is, you wouldn't want to pay thousands of dollars for a robot to make you a sandwich or clean your toilet because those are things you can easily do yourself for free. And there's no economy of scale to offset the cost, because it's just doing it for only you.

How about "get your bot to build my microprocessors" or "get your bot to route my package through a warehouse" or "get your bot to weld a car"? There you can get robots to do amazing things, and they can cost whatever the company wants to charge because those robots are making money. If a robot can build a modern CPU, it could clean your toilet. That's not the problem.

Even still, there are household robots that are fairly common. I've got one running upstairs right now. It's called a Roomba.

View on HN

"Of course a robot could do those things. Those are very easy things to do. "

Sorry, but no. Very, very hard.

+ Find my fridge + Open it + Find the meat + Find the bread + Open the mayo + Get a knife + Spread it

etc. etc. etc.

Even if you do it 'the machines way' it's extremely difficult.

No way. Not for quite a while.

Much harder than 'driving a car'.

Without proving it directly - let me do it indirectly:

Factories would love to have robots with such intelligence and dexterity. It's definitely worth '10's of thousands of dollars' per machine to do that.

If machines could do that ... even Chinese labourers would be out of a job.

We can barely make robotic arms that 'hold stuff' of any weight reasonably without getting out of balance.

Go ahead and look at industrial grade robot-arms - you'll see these powerful steel machines - max load: 5 lbs (!).

McDonald's would be having robots prep your burgers - even ina a controlled/machine setting - and they don't do it - because it's a ways off.

There are so many issues with that one, it's a long way off.

If you get to do it "the machines way", then you get this - https://www.youtube.com/watch?v=aPTwpLIpDWE

I agree with the idea that we are 20 years away from a walking talking robot. One that could find the fridge, figure out which mayo is the good one, and which mayo has gone bad. The get the right ham, not the crap the kids like, and find the knives. Where are all the knives? And then put the stuff away, so nobody complains later....

I own 2 welding robots.(Miller-Panasonic Prefab Welding Cell) They are time consuming to program, and you do have to 'touch up' the program all the time. They are both 7 axis machines, with the capability of running another 13 axis each. The complexity of a setup is much much higher than a self driving car, which has go/stop/left/right as possible inputs. A self drive car is only 2 axis of movement.

In order to deal with 900 types of sandwiches in another factory, humans are used, not machines . http://www.mirror.co.uk/news/uk-news/sandwich-factory-worker... . They don't do it because they like paying humans. It is too complex to handle the high mix of products (900 sandwich types) for automation.

> If you get to do it "the machines way", then you get this - https://www.youtube.com/watch?v=aPTwpLIpDWE

Those are some of the worst sandwiches I've ever seen – which confirms your point, I suppose.

Are you saying that everything in the food industry is done by humans? Watch an episode of Unwrapped on Food Network and you'll see that everything you've just said is false. Robots make a lot more of your food than you might think.

Again, they just don't do it in a "human" way. They do it like robots do it. Driving a car is infinitely harder because you have to do it the human way.

Again, you're thinking of making an android: something that resembles a human and has fingers and hands and would spread butter with a knife. That's a far cry from "a robot that makes a sandwich". Those have existed for longer than the Internet has.

Making a sandwich or cleaning a toilet is certainly not an easy thing to do. Getting a robot that can understand a scene, any scene and plan motions within it has been very difficult to do. Far more difficult than playing Go at a superhuman level for example. This is known by various names, with one common being the Moravec Paradox. The problem of energy storage further multiplies this difficulty.

I think we're getting confused on the definition of how the robot accomplishes its task. To clean a toilet like a human would, yeah that's difficult. To clean a toilet like a robot would, it's not. A toilet cleaning robot would have a spinning brush like a drill and would circle the bowl of the toilet a couple of times. Really, really easy thing to do. If a Roomba can clean my entire house, not fall down the stairs, and find its way back to its dock by itself, you can put it on stilts and attach a spinning brush to the bottom and let it find the toilet.

Likewise for a sandwich, if the meat and cheese and bread are all in robot-compatible forms, it's just layering them together. Not much more complicated than a printer that can pull from multiple paper trays. You can even combine the two for maximum efficiency.

What's hard is making a robot behave like a human, because even humans struggle at behaving like a human. IMO there's no purpose to making a robot have the same physical constraints as a human. Our bodies are pretty shitty when you think about it. Robots can be much more elegant.

If you don't think there are sandwich-making robots that actually exist, I'd wonder how you think gas station ham sandwiches are made. Hint: it's a conveyor belt that drops bread, then cheese, then ham, then more bread. That's a robot that makes sandwiches. I'd sell you one for $1m, but you'd probably say "I can make my own for free, thank you!"

Here you go https://www.youtube.com/watch?v=aPTwpLIpDWE&t=8s

> A toilet cleaning robot would have a spinning brush like a drill and would circle the bowl of the toilet a couple of times.

But that robot can't refill its consumables, it can't mop up the floor, and it can't deal with the mess after your semi-demented grandpa missed half the bowl. This solution is useful for a large rest stop with a dozen toilettes and thousands of customers per day (and employees that do the above tasks), and utterly useless for your own home.

> IMO there's no purpose to making a robot have the same physical constraints as a human. Our bodies are pretty shitty when you think about it. Robots can be much more elegant.

Our bodies have a huge advantage: they are incredibly flexible, and surprisingly strong for their dexterity. No robot we have ever built even comes close. You can have lots of robots in your home (Roomba, toilet bowl cleaner, dish washer, sandwich maker, a kitchen machine, uuh, is there anything else you can automate?) and you'll still need a human doing a lot of work. Or you can have a single human working half-time in your home, and you don't need a single robot. Give me a robot that can do 90% of all household chores, and I'm sold.

But now you're getting into scope creep. What was asked for was a robot that could clean toilets, or a robot that could make a sandwich.

Wasn't the original discussion about robots replacing human labor? A single human can do those things.

I think it's another discussion whether we can make robots who each do one thing well. Roomba, etc.

Besides the space constraints with multiple robots, I think it's clear there are significant advantages in having just 1 entity that can perform all the required cleaning tasks. Much like single-purpose hardware has given way (mostly) to software, except in areas with interesting requirements (GPu etc.)

And a self-cleaning toilet with that brush and software would be more effective and cheaper...

The comment which originated this subthread was about telling a bot to get a sandwich and clean the toilet. This implies a single bot and the verbiage implies an agent with strong autonomy.

The disadvantage of the specialized machines you speak of is their limited to no adaptability, inflexibility to novel situations and obsolescence upon changing environments. There's need for actually smart appliances (which is what the kind of robots you spoke would be) but they'd be best suited to easily circumscribable tasks. Not what many have in mind when they think robots.

Reply on news.ycombinator.com