I think the thing they want to communicate here is the training method. Rather than a pick and place tool path, this is just trained on videos of someone using a coffee machine, and imitating that.
Which… now that I think about it… is probably drawing a skeleton over top of the video Kinect-style, and… drawing a tool path…
But at least it seems to understand when it’s made a mistake, which is interesting.
Right I mean, for B2B industrial robotic arm applications this might actually be a big improvement / breakthrough.
I just find it annoying how many of these AI companies are pumping goofy B2C applications.
Like they make it look as though they are building a breakthrough humanoid robotic assistant or something, but really that is still very very far away and not at all related to the "video trained robotic arm" breakthrough they are trying to communicate.
I don't find this particularly impressive as well. Anything except a pod machine would likely present the robot with tasks it can't handle like e.g. handling actual beans or ground coffee and filling water. And most household tasks would often involve this kind of task that is not entirely straightforward as your house and its contents aren't designed with limited robots in mind.
This Keurig queued up perfectly use case is fundamentally unimpressive because I could expect my 5 year old nephew to pass it with 1 minute of verbal direction or watching me do it once, not 10 hours of deep learning.
Now expand the Keurig use case to the real world.
Present it with a machine:
* filled / not filled with water
* used pod bin empty / used pod bin filled
* pods available / pods in a box on counter / pods in a box in the cabinet
* cup on counter / cup in cabinet / cup in dishwasher / cup in sink
We now have ~50 permutations of even the simple Keurig use case and let me know how many years of training to accomplish even this. All of this to make fairly bad coffee.
We haven't even gotten into your example of actually making real coffee.
A general purpose robot can learn new tasks in a modest amount of time.
As you argue, fluidly chaining sequences of learned tasks together is tricky, but current AI technology is capable of solving nearly all those problems within a handful of years.
With centralized learning labs, over-the-air updates, and a 2x or 3x improvement in execution speed of precision tasks, I think this type of solution is going to be ubiquitous within the decade.
From 10 hours of video, it learned to pick up a pod carefully placed on a counter in front of it, insert said pod into a Keurig machine also carefully placed on the counter in front of it, and push the button.
From the hype, I was expecting it to pass the full Woz test:
1. Walk into a strange house and find the kitchen.
2. Find whatever coffee is available, including knowing likely places to look. Choose the best among multiple options based on the best available information about the drinker's likely preferences.
3. Find whatever coffee maker is available, get it out without screwing anything else up, assess its condition, clean it if-and-only-if necessary, plug it in, figure out its controls.
4. Similarly find filters or other supplies if necessary.
5. Grind the coffee if necessary to match the coffee maker and the desired results.
6. "Make the coffee" in the sense these guys are trying to claim is a breakthrough.
7. Serve it, or at least put it it in a cup and tell the user.
8. Put everything away and clean up
9. Take any available feedback on making it better next time.
Did a quick search for anything at all with more detail than the twitter post, and couldn’t find anything. They sure do have a ton of hip marketing image. But they also seem to have some clued up engineers. So who really knows
To me, there's only one metric that matters for humanoid robots: Can it earn its keep?
A $30K robot with 10 years useful lifetime, and a 10 year 5% mortgage is $10.40 per day. Minimum wages range from $7 to $15, so the US national average would be around that. If your robot can do 1 hour of minimum wage labour per day, people will buy it.
This is a pretty low bar. Light cleaning (dusting, wiping down surfaces, vacuum), loading the dishwasher, doing laundry, tidying up after kids, etc. These tasks don't demand AGI, agility, adeptness, speed, accuracy, etc.
Do keep in mind that a comparison like that doesn’t suppose the null case, where someone dusts their own home for free, loads the dishwasher, and trains the kid AI to clean up their own messes with thoughtfully deployed apple-sauce-snack-based positive reenforcement training.
I do agree these tasks don’t demand AGI. I don’t really want my “robot that walks around and does 11$ worth of tasks I don’t want to deal with” to feel emotions.
Disappointed. The robot t is not "making coffee". It is feeding a capsule to a machine and pressing a button.
I expected the robot would be grinding beans, filling the coffee holder, tamping, fitting the holder to a steam machine and pulling an espresso for a certain time.
I could even understand fitting a coffee filter to a holder, dropping ground coffee and dripping hot water until a certain level is reached.
Those things require a lot more decisions than "open machine, drop capsule, close machine, press button".
Why does every hype-funded general purpose robot (Tesla bot, this robot) NEED to be shaped like a human. I get that our built environment is designed for humans, and some sci-fi loving part of me loves it.
But it feels like theatrics. Boston dynamics makes utilitarian robots, that all have a seemingly purpose built design to them. Even the bipedal + 2 arms + a head robot Atlas, isn’t forced to constrain itself to human proportions. By comparison this robot seems pretty unstable. I wonder if that has to do with humans and robots being fundamentally made of different materials with very different properties.
I probably answered my own question by calling it “hype-funded”.
One reason could be that they train the robot by looking at a human. Then they point a camera at the robot when they perform the task. So if the robot looks like a human, then it is easier to copy the movements of the human. Perhaps. Just guessing. I haven't read the article because I dislike the format.
> I probably answered my own question by calling it “hype-based”.
No, I think you answered your question here:
> I get that our built environment is designed for humans
I don't want a purpose built robot that does one task extremely well, I want a general purpose robot that does a huge variety of tasks well-enough. Make me a coffee, then do laundry, then rake some leaves. Coffee served in the wrong mug? Clothes laid flat and not folded because it can't actually fold? Random leaves still on the ground? It's fine, it's good enough! It still saved me a bunch of time.
Remember, these are aimed at car-level prices. $30k for a machine that lasts 10 years means you're paying $10/day, you do not have to be very good to earn your keep at that price.
I understand where you’re coming from, but I’m more so pushing back on the added cost and complexity of human elements NOT required by the built environment, or are even detrimental to it even acting like a human.
There are not many tasks that require a neck for example. Or a whole head really. This robot feels like it’s designed with a shape in mind first.
The dexterity and wobblyness is actually pretty unimpressive in this example, compared to Atlas, which absolutely feels closer to the full range of human dexterity and physical ability.
Atlas is still general purpose. It’s just built with human tasks in mind first, rather than a human shape.
I get that, but this thing in particular does not have a neck. It has a metal stand that's shaped to look like a neck, but it is a static, fixed piece of metal, it doesn't rotate, tilt, nod or anything of the sort. It's a stylistic element that's not relevant to its functionality.
Atlas is a wildly overbuilt thing that's simply not relevant to majority of human tasks. Doing parkour is very impressive, but nobody needs to do backflips to clean the kitchen. The average human has far far fewer capabilities than an idealized 99.99th percentile human. The real world is accommodating of people with various disabilities and limitations, you do not need to be an olympic athlete to live a productive life.
>but I’m more so pushing back on the added cost and complexity of human elements NOT required by the built environment
Am I understanding you correctly that you would want something that still had cameras at head height, but simply didn't have the head form factor? And perhaps had 4 arms for some tasks that it would benefit from instead of being limited to two arms?
If so, how many environments are you going to build robots for? And how does your total overall build cost increase with each different model you build?
Doesn't matter if it's simper in the end-design if getting there ends up costing you as much as it would to build a general purpose design in the first place.
Sure, its an interesting goal but it does not seem anywhere near a this-decade deliverable.
I'm also less convinced about having all my chores done half-a**sed for $30k up front cost. A $30k investment every 10 years is either going to cost the buyer interest when they finance the purchase, or foregone returns/interest if they keep the lump sum invested. So the true cost is more like $45k. Further, it's now a device you own and are responsible for maintenance/charging/etc I am sure. So now we are talking I dunno, maybe $50k.
And at that price .. $30-50k per decade, you could outsource a subset of your chores to specialists to do 100% of the job, and leave yourself with the subset you don't mind.
$3k is in the ballpark for an annual yard work budget even in HCOL areas. For $5k annually you could probably outsource your laundry entirely with some light housecleaning.
Or you know, find the neighborhood teen (or your own) and see how many $20s it takes to get them to clear your driveway / do your yardwork.
I've been very squeezed for time the last couple of years and have tried to buy my way out of this situation, which is one of the reasons I'm so bullish on humanoid house robots.
I hired landscapers one year for over $3K, they did well but they came once a month. It was way too expensive for what they did. I tried hiring a guy on an ad-hoc basis, he did a bunch of half-assed work for $30/hr, then disappeared leaving a bunch of garbage in my yard. Cleaners are available for an $25+/hr on freelance sites. I have someone coming in right now for $30/hr, and they're ok, but they only do 4-5hr chunks and so the house has to get noticeably dirty for them to be worth it. No one will come for 1hr/day and do laundry, they have transit overheard, scheduling, etc. Snow clearing is $300+ per season, but they don't show up for anything less than 3", which happens only ~5 times per season, and they'll only do it within 24hrs (ie, you're not going to get your snow cleared when you need it, because everyone needs their snow cleared after a huge snowfall, you might actually be stuck in your house). I spent a bunch of time finding a student do clear snow and he came once and then ghosted me.
I could keep ranting, but the point is simple: the reality is people are hard to find, more expensive and less reliable than your post implies. There is a lot of overhead and the end result is that a bunch of work that should get done, simply does not get done.
A $30,000 10 year mortgage at 6% is $11/day, $4000/year. Charging costs are negligible, like a couple of cents per day. My EV moves 2000kg 100km for $1.50, moving clothes and brushes around the house is nowhere near that. Any competitive company will also provide regular OTA updates that makes your robot more capable over time.
The bar for these humanoids to succeed is very low.
The bar for humanoids to succeed may seem low, but think about how we are like 10 years into "FSD next year for sure". In that case, the car already exists! All the computer needs to do is steer/brake an existing automative frame using vision & compute.
A humanoid robot needs to essentially deliver the navigation ability that FSD has failed to do, as well as a new robotic humanoid frame. The frame hardware & nav software needs to navigate tighter spaces, with tighter precision (fractions of inches rather than feet) in closer contact with flesh&bloods easily damaged humans (within inches of us rather than multiple feet away on a sidewalk).
Think about yard work. Even a half decent yard service has a half dozen bladed tools they wield around your yard. What is your risk appetite for a robot to wield those autonomously around children/pets/neighbors/wildlife/you.
Most industrial robots disable themselves if humans are anywhere nearby and a lot of accidents happen when that system fails. It is going to take a lot of confidence to put robots into homes.
Efficiency wins out over sci fi coolness, every time. You’re totally right IMO. When it comes to most tasks, I normally care about the task being done, and done to some acceptable level of quality. Hiring someone gives me that for a fraction of the cost, so sorry robots, but it looks like humans are taking your yard work jobs.
Bipedal robots will be a common thing in industrial/commercial settings within the decade. There is just such a huge financial incentive to create bipedal robots to take over human tasks. $100k for a robot that can work in a warehouse and lasts 10 years working 24/7 is far less than what you’d have to pay two shifts of minimum wage workers.
Again, my point I made in other parts of thread - B2B use cases, yes. B2C use cases, no.
Factories have been designed since the days of Henry Ford's assembly line to treat humans like robots. We have increasingly automated/mechanized the assembly line, introducing robotics of various sorts. So sure, a limited-functionality bipedal robot that costs less than high cost union labor, doesn't go on strike, doesn't take smoke breaks, and doesn't need to eat/rest/sleep is attractive.
But as a general purpose home assistant / personal housekeeper we are far far away, and especially for anything remotely affordable even for FAANG types.
This x1000. My sense is that because of the historical development of how we talked about AI in constant comparison to human intelligence, along with sci-fi dreams, there's been a long-standing implicit value assumption that we necessarily want robots that emulate humans, but I'm highly skeptical this is actually desirable. It's quite plausible that robots that are just robots and don't waste effort on human emulation are actually far more beneficial societally speaking.
I would think of robots as tools.
And in that sense, over time humans tools have gotten more advanced and specialized.
Think about the various power tools & garden/yard tools a median DIY homeowner accumulates today vs 50 years ago.
So to me, robots make more sense in purpose driven forms than in some humanoid replica attempt. The humanoid walking form factor is far harder to accomplish, and in many senses possibly less useful.
I think what a lot of people want, again influenced by sci-fi, is a general-purpose robot that can do everything a human can do and preferably other things beyond that. One form factor that can do everything we do is obviously the way we humans look hence the inclination to design robots that look like us as well.
I could have multiple robots, one for vacuuming, one for doing the dishes, one for gardening, cleaning, renovating, etc. But, if I had one to do all of it any of the purpose-driven forms is likely to fail while the humanoid one _could_ succeed.
Humanoid robots strike me as a FSD/AGI level of overshoot. Sure a humanoid robot that actually worked would be able to accomplish everything a human could physically because of the way interior spaces are designed. But focussing on that deliverable may miss a lot of available use case solutions.
For example robotic vacuums and lawn mowers have been around, pretty cheap, and pretty good for some time.
There's lots of home chores that a wheeled robovac with telescoping arms & really good vision could accomplish very well (empty/load dishwasher to/from sink/cabinets, light kitchen work, dusting, sweeping, etc). To deliver this in humanoid form you need to solve all the same vision/arm articulation problems, PLUS all the self balancing/walking/look&feel stuff of a humanoid.
There's actually stuff a robot could do better than me if it had a telescoping arm instead of being humanoid look&feel focussed (dusting / reaching high places etc).
To me it conflates 2 problems and overcomplicates to focus on the humanoid form factor at this stage.
And look at the evolution of some tools - cars do not really look like horse drawn carriages for example. Helicopters don't look like flying cars.
Agreed with caveats. A good general robot would still probably have similar mass to a human, appendages with similar reach, and so on. A horse carriage and a modern car have similar sizes, can take similar amounts of cargo and seating, and take up similar space on the road.
So a good generic bot needn’t be an android, but in that ballpark.
Boston Dynamics basically just makes proof-of-concept robots, and then they pre-code demos for videos.
If you have a non-humanoid robot, like a Kuka arm, it's a "sharp tool" which is overfit to one specific range of tasks. It's never going to be able to do anything else, and if you wanted to add other tasks, you'd have to overhaul it or create some adapter.
A humanoid robot basically needs no adapter for any common task, in that every common task is already performed by humans, who all have the same components. The benefit is that the range of tasks available to the form factor is massive, versus a Boston Dynamics robot that just picks up and moves packages and can only do that. If the humanoid robot is not only far cheaper, but it can do that tasks, and thousands of others, then that's the better form factor..
When we talk about the 'Uncanny Valley', we talk about things that are very humanoid, but not quite humanoid enough.
But the obvious target market for humanoid robots is retail and hospitality, where there's a high degree of human interaction. In that case, a lot of 'humanness' is needed to suggest to customers that 'yes, you can talk to the robot, and it will do things'. The design of this robot[0], on the other hand, is very un-human (poorly-drawn face aside), telling customers that it has a specific job that doesn't involve them.
But doesn’t that make the value in these human-y robots, simply impressing humans? I’m fine with theatrics and awe being the main purpose, but it’s really just the front end to get humans speaking to a hunk of metal without feeling weird. What difference does it make in utility if it was a screen with some 3D model talking back at you? That’s a lot easier to make feel “human”. And now you’ve freed up the design of the robot behind the scenes that actually does the task to be shaped in whatever format is useful.
Is the retail space going to get similar value from the robot once the initial “oh wow robot cool” factor wears off?
I don't know that replacing $15/hr at-will labor with a $50k machine that can be vandalized/stolen/etc is going to have a great cost-benefit for most US retailers...
Perhaps because they can reuse the robots for a thousand types of tasks so they don't have to reinvent the wheel every time? Also much cheaper that way.
You’re conflating what I’m saying about human-styled robots, and the reality of humanoid robots.
We’ve been making bipedal, two armed, dexterous robots for a while now, to varying levels of success. They don’t tend to obsess over the slick human look though, since it makes wobbly unbalanced machines, like the one in this video.
Compare to Boston dynamics atlas [0], whose balance and dexterity is almost a bit scary. Is that run a result after many failed runs, yes. Is this video from Figure the result after many failed runs, I would also assume yes. I am much more impressed by what Atlas does though.
+1 but even Boston Dynamics stuff is too fancy and animal shaped. Here is a great example of utilitarian robot design that looks much cheaper and more efficient than bi/quadrupeds: https://www.youtube.com/watch?v=ox7AcAWWxyI
Aren't humans rather inefficient? Long gestation cycle with more complications than many other species. The offspring require long time to be self-reliant at any level. And time to puberty is also significant...
Seems like that even for breeding we could do lot better...
My opinion is towards the end of this decade. Hardware is mostly there - the collection of actuators, metal bits, batteries, cameras and inference chips means you can probably build this today at scale for the price of mid-range car, but of course it wouldn't do anything useful. But given the rate of progress in AI over the last few years and the increasing rate of investment in the field, I'd be shocked if another 6 years of progress would amount to nothing.
I mean "learned to make coffee" is a bit exaggerated? It just placed a capsule in the machine and pressed a button? Possibly the coffee machine has contributed much more in making that coffee and doesn't even have an AI system :D
The robot is not "making coffee".
It is placed in front of a Keurig, coffee cup already in place, water already filled, powered on.. with a pod in front.
It is doing essentially 3 things - pick&place pod, secure latch, press go.