Robot teaches itself to make coffee

Andy

1 Like

You beat me to it! Well done!

The important news is that the robot learned to brew coffee simply by watching humans doing it. In previous posts I’ve mentioned how children learn but they do acquire language. Animal cubs don’t have human level language yet they still learn by watching. What this is telling me is that non rational neural nets have the emergent property of learning. All that is needed is the requisite complexity.

Not to worry, Humanoid AI Robots deniers will say that until the robots can smell the coffee and determine of it’s for for human consumption they not really Humanoid Robots.

The Captain

3 Likes

Yes, if that is true it is a game changer. Any task could be taught easily and then when one robot learns it the rest will know what that robot knows via a software update. It could be done wirelessly and instantaneously.

Andy

3 Likes

I didn’t watch the video. Did they mention how did the coffee taste? Good coffee is more than coffee, machine, filter and water. So interesting what is happening in tech. I have already begun to follow robot stocks…doc

Actually, no. The VW dealer I take my car to got a new gadget last year. I push a button, it heats the water, grinds up some beans, brews, and pours. Best coffee around. The difference is the beans. First sample of it’s work I tasted was like drinking battery acid. That was from the beans that came with the machine. Once that supply was exhausted, they went back to their usual beans, from Zingerman’s. Excellent coffee.

Steve

2 Likes

Hi Doc, It was a robot putting a pod into a keurig. So Maybe not for someone who is a connoisseur.

Andy

3 Likes

LOL, thanks Andy. That’s even more clear…doc

Don’t expect instant gratification but finding out that robots can learn visually is a big step towards further developments just as large language models are. Once we can tie all the threads together we’ll have real progress.

One of the things you missed by not watching the video is that Google’s Mobile ALOHA has a lot of interest from many large corporations which somewhat reduces Tesla’s advantage of using Optimus in-house if Google and the interested parties find a good way to collaborate.

The Captain

5 Likes

I am not a coffee aficionado, but I’ve been told by those who are that the most important thing is the beans, the roast, the grind, and the temperature of the water as it goes through the grinds. Maybe I’m forgetting something else, but I think those were the highlights.

All I know is that Dunkin coffee is terrible, Starbucks is just okay, and when I buy from a coffee shop, cold brew is the best tasting to me.

1 Like

Also one thing that the video showed you was that the robot first made a mistake putting the pod in the keurig and was able to self correct itself. Maybe has the ability to solve problems?

Andy

3 Likes

And on the flip side, and article on Why It’s So Hard for a Robot to Straighten a Candle Wick:

https://archive.is/iSuZz

The issues raised in that article aren’t caused by the form factor of conventional robots, BTW. It’s just really hard to mechanize the job of manipulating an off-center candle wick. Again, not that any of these humanoid robots have to be able to do that specific job to be useful - just to point out that many (maybe most?) of the factory jobs that are presently not mechanized are still being done by humans for reasons unrelated to the form factor of the robot.

1 Like

There are robots and there are robots. In the article they don’t mention AI trained humanoid robots because they didn’t exist in the time frame they are reporting on. We are talking about AI trained humanoid robots so the comparison is Apples vs. Aardvarks.

L-i-m-p candle wicks present the same difficulty as l-i-m-p automotive wire harnesses. Tesla solved the problem by redesigning the wire harnesses. Maybe the candle makers could use Viagra to solve the problem. It works on l-i-m-p human wicks! :star_struck:

The Captain

Aardvarks

have wick like tongues to get at the ants the love to eat!

To make sure there is no confusion between the similar looking beasts as there is between robots and robots, here is an article to point out the differences

Interesting: Aardvarks and Anteaters: What's the Difference?.

Neither is like apples which are fruits. Just making sure…

An error occurred: Sorry, you can’t post that. Not allowed: L-i-m-p, l-i-m-p.

3 Likes

The article is from yesterday. How would a humanoid robot solve their problem in a way that couldn’t be solved with a robot with a different form factor? You’re not going to get better motor control with a humanoid hand holding pliers than a robot that has pliers as part of its mechanism.

Albaby, the AI trained robot trains itself to to the job. That is huge. It is like a person who only has one good arm. It doesn’t matter that they do not have the same form factor as someone who has two good arms if they can train themselves to do the same job with only one good arm. It is nice to have the very best motor skills but is it necessary to have the very best motor skills? I think not.

Andy

1 Like

But if you have an AI that can train itself to the job, why put it in a humanoid body for that factory? Why not just put it into a mechanism that has, say, 5 or 6 ‘arms’ that just end in the type of pliers that are necessary to straighten the candle wick - and mount it above the conveyor belt so that it doesn’t take up any room on the floor?

That’s the beauty of whatever AI (or any other “brain”) you might develop. You can put it into any type of external body you want. Why put it one that isn’t designed for the job you need it to do, in a factory setting? Why give it legs if it’s never ever going to move from its wick-straightening spot?

I explained this before. Now if you don’t agree please let me know. The reason they are putting it in a humanoid form is because the work place is made for the humanoid form. Humans may need to take the place of the robot for repair work or to fix some problem that may come up. So the work place still needs to be designed for humans. Not only to keep it convenient, but the highest priority is safety. Also the humanoid form, if filling in jobs that have been formed by humans, can fill in anywhere. If they did it the way you stated each robot would only be able to perform that specific job. The way they are planning it, the robot can fill in for any job. So one day a candle maker the next day a car mechanic, etc. Now if you watched the video and if it is true that the robots can teach themselves, either by somebody controlling them until they learn, or by watching. Then it would be possible for them to do any job. Not only that they said any robot that learns a job can teach any other robot. Not like a human is taught but OTA transfer. Almost instantaneous. I don’t know if that is true…yet. But if it is, that is a real game changer.

Andy

4 Likes

Not all workplaces. Many factories are made for both the humanoid and robotic forms. Most robots in factories don’t need to use any of the elements that are made for the human form. There’s no benefit to a humanoid form for a spot-welding robot. Why give it legs, and a hand to hold a spot-welder, when it never needs to move or do anything but spot-weld?

And some aspects of the factory are made to accommodate the limits of the human form. Humans only have two hands, have arms that are only so long, are only so tall, require tools to do most jobs…etc. Robots don’t need to replicate those limits, and will often be much better at their jobs if they don’t have those limits.

That’s not even close to true. Any more than it’s true today that if a robot can be built to do one job, it is possible to build a robot to do any job. Or that if AI can do one thing, it is possible for AI to do any thing.

Many human jobs are amenable to being mechanized, because the actions required fit well within what robots (AI or otherwise) can do. Manipulating rigid, uniform, relatively non-fragile and completely non-variable objects (like putting a standard-size Keurig pod in a coffeemaker and pushing a button) is a job that you could program a robot to do twenty years ago. Many (most) jobs like that were replaced by machines long ago. What remains are jobs that are more difficult to mechanize.

The AI tech we have now

1 Like

I think you can think of a benefit albaby. Why do I put tools in my shop on wheels? I mean I could just bolt the tools to the floor and go to that spot every time I need to use it but I put them on wheels why? Well maybe I need that tool for a job in the other part of the shop, or maybe I want more floor space so the tools all get shoved to one end of the shop. Now I could buy the same tool and put it in a different spot in the shop but why? Making something mobile has many advantages, but having something mobile that can do many jobs is even better. Now I have a welder, that turns itself into any job I need. All you have is a static robot that can only do one job. I can run my robot all day and then rent it out to someone else and have it go clean their house at night therefore making me more money.

Why would you want to limit your robot?

All of what your saying does not take into any aspect of safety for humans either. If a human never had to go on the factory floor that might work but as soon as you put a human on the floor you have to design it for safety.

I also don’t think you are taking into account that the human formed robot is definitely coming, in fact it’s here.

Did you even watch the video?

I made it clear I do not know if it is true but that is what they are claiming in the video.

Andy

1 Like

Because you don’t have an assembly line. Part of your shop’s operation is moving the tools to where the work is being done, rather than having large amounts of the product move through the shop to where each step in the process is done.

For the same reason that you have lots of different tools in your shop, rather than using a Swiss Army knife or a Leatherman for every job. Occasionally, flexibility is valuable - but in most cases, and certainly most cases in large-scale manufacturing, the right tool for the job is more important than flexibility. If my robot is working one station in the assembly line, I want it to be optimized to perform that task.

Yes. Lots of people who are enthusiastic about some new tech will say lots of overly optimistic things about that new tech. Whether it’s the Segway, or blockchain, or the MetaVerse, or what have you…this New Thing is almost unlimited in what it can do!

Except it’s not. There are limits and constraints and inefficiencies that don’t get discussed in the enthusiasts’ videos, but are obvious if you step back and think about it for a few minutes. We know that language and image AI - which is the new hotness - is good at doing some things, terrible at doing other things, is average at still more things, and is just irrelevant to many other things. There’s no reason to believe that as those new large data techniques are applied to robotics that it will be any different.

2 Likes

The reason I do not have an assembly line is because I do not have the space. Space is waste. The less the floor plan, the less the heating and cooling, the less the taxes, etc. A mobile tool makes life so much easier also. Take a part washing station, put it on wheels, full of 20 gallons of fluid that is 200 pounds wheeled against the wall waiting to be used.

That is incorrect Albaby, I have different tools because I can’t get one to do all the jobs. If I can get a tool that does it all , that is exactly what I buy. My welder does Tig, Mig, and Stick. I could have bought three but I bought one that does it all.

Ok, well if you do not think it is possible that the people in the video are correct than this discussion is moot. Because that really is what the discussion is about. We will have to agree to disagree.

Andy

2 Likes