Robots that can walk are now striding to market
economist.com
economist.com
Much of this is a mechanical design problem. You need to be able to generate large forces from a small package, but not all the time. Hydraulics can do it, but then you need a big pump and power source, like the older Boston Dynamics machines. Until recently, all-electric didn't have the power density. Now it does. Which is why quadrotor drones work.
Small polyphase permanent magnet motors ("brushless DC") have taken over. The controllers to drive them at variable speed are now cheap. No brushes, no arcing, no magnet demagnetization. The remaining limit on motor power is cooling.
It's too bad Schaft Robotics was liquidated. They were doing well on humanoids. Then Google bought them, trashed them, and dumped the remains. They were into small liquid-cooled motors. That's a way to get huge torques when you need them.
Reduction gears are always a problem. If they're big enough not to snap gear teeth on shock loads, they're too heavy. Better materials for toothed belts have helped. Those have more space to absorb a shock load. Researchers like harmonic drives, because they can put a big reduction in a small package, but harmonic drives are especially vulnerable to broken gear teeth on overload.
Direct drive, without gears or belts, would be nice. The advantage of direct drive is, as someone said a century ago in the context of electric locomotives, "You cannot strip the teeth of a magnetic field." Washing machine motors are often direct drive today. The electronics is cheaper and more rugged than a gearbox. It's encouraging seeing that happen in a mass-market product.
There's a lot of engineering like that which needs to be done to get mobile robots down to the price of a non-luxury car.
This seems to be a too oft repeated refrain. Google has become such a mire of intra-office politics that being acquired by them is almost a guaranteed death sentence.
They sure fulfilled that contract.
They continued doing this during the war, ... and during the "trial run" holocaust by the German and Austrian Youth Services, then the continuation of that trial run by the psychiatric institutions ... and then with the actual Holocaust. They used a Polish subsidiary, run by a Nazi (owned 90% by IBM), in case you were wondering how they did that without being convicted of treason.
The last time I looked into this it didn't look like that much torque was available in a small form factor. I found it rather surprising because larger scale industrial robot arms appear to far exceed any biological arm in both strength and speed.
It doesn't need to lift heavy things or do many tasks himself because it could use dumb tools/machines to do that just like humans use forklifts.
That's why I'm saying being smart/ingeligent is more important than how much it can lift or its battery autonomy.
Plenty of core muscles involved.
This means that the force of your fingers is generated by a muscle roughly the size of a cola bottle, the force of your lower arm is generated by between 2 and 3 stacked cans of cola, and the force of your entire arm is generated by a muscle the size of a decent shoebox.
So while muscles are pretty good, they're worse than our best motors. Or at least, if you can provide them with the power they need they are. And you think this is not the case because your body hides pretty sizeable muscles in your torso where everybody thinks you have almost only organs and fat, when in fact about half is muscle.
Most robotic arms I've seen appear to have a motor located at the elbow joint that drives the equivalent of the lower arm either directly or through a gear box.
If cables and pulleys give better torque without loss of speed why aren't they in more widespread use in robotic arms ?
Because it requires more parts, so there's more to break, it's more complex, it loses accuracy, and it fundamentally limits range of motion.
I've seen direct drive SCARA robots with a big pancake direct drive motor on top. Those are used for vertical assembly, for designs where a straight-down move puts the part in place. The arm moves in X and Y, then there's a vertical piston-like actuator that moves in Z, and a rotation axis around Z.
Something that doesn't seem to have caught on is putting tiny cell phone cameras in the end effector, so you can fine-tune positioning.
When you see surgeons, who have loads of theory about precision movements you will quickly find that they rely on feedback rather precision for their movements. For example, they cut along the "linea alba", not by knowing where it is and positioning the knife exactly, but by simply slicing down starting at a good enough position, knowing the knife will follow the linea alba because it's attracted to it. Which is a fancy way of saying that it's a force feedback problem, not a positioning problem.
I can hit a 2mm square target reliably with my an instrument held by my fingers (and my hand resting or held in place) and, say, a 2cm target with my hand. I get that I'm probably not surgeon material, but let's say they can double or triple that accuracy. 10x is out of range of any human without robotic help.
Robots are expected to do that with three numbers behind the decimal point and to keep that position, while carrying their load (varies from 200gr to 20ton). There is no way a human can do that, not even with 200gr, to an accuracy of a centimeter. And even trying will be so stressful you wouldn't be able to hold it for even a minute. A robot can hold it for 10 years.
If a human family member holds something like a baseball bat, or a butcher's knife within arms length of you, you can feel safe around them, and be sure they control their motions.
Contrarily, having an industrial robot with something like a screwdriver at the height of your head and within a meter of distance is just scary.
https://www.youtube.com/watch?v=s6_azdBnAlU
His takeaway was that the big use cases currently are around LIDAR and mapping, e.g. tracking construction progress, in irregular / uneven terrain.
I'm less convinced that walking robots are useful for some of the use cases described in the Economist article, like moving boxes in a warehouse. After all, warehouses are generally built to facilitate wheeled vehicles (forklifts) already...
Legs allow you to climb, ie a ladder, very steep stairs ( more than 45 degrees), a tree, or an extremely steep rough slope (think: mountain).
A tracked vehicle or large-wheeled vehicle can go almost anywhere a legged robot can go with a large efficiency and cost advantage.
However, you can run into space constraints. Bipedal robots are good for navigating human-scale environments while still having a tall and stable robot.
But generally, tracks or big wheels are way cheaper and more efficient and can do the same thing.
but yes, bigger wheels and tracks probably cover a greater percentage of the commercially viable landscape.
also, your point is why i mentioned pearls--not all tissue needs to be activated or enervated similarly, or at all.
we should have no doubt that biology would find a way given enough time.
Actually, I suggest its for exactly that reason.
Rotary motion is sufficiently more efficient that if there was a way to evolve it, it probably would have.
Rotary motion has lots of difficult issues: how do you power it, how do you communicate with it, how do you lock it in place, how do you reverse it?
None of these are problems for stabilized linear actuators (aka muscles).
legged locomotion has evolved to be quite efficient in certain ways too. human legs, for instance, evolved for walking/jogging long distances efficiently (we have 2 gait-dependent locomotory efficiency points at two different speeds). but we can also climb and run and kick and hop and swim and more when required, unlike rollers.
Unlikely. Something that has the possibility to be so staggeringly more efficient would have found some niche even in present day. At least one spider makes itself into a wheel to roll down sand dunes faster than a predatory wasp can follow.
When something as complicated and calorically expensive as vision can evolve multiple times, the fact that rotary power hasn't evolved says that there is a block somewhere.
And, rotary power has evolved in some flagellar creatures. However, what evolved was a cellular ratchet. The system can go one direction only, and there was effectively no fine control for exactly the reasons I mentioned.
> for a potential analogue/precursor, consider the rotary joints we have in our bodies.
And those joints have complicated musculature and are relatively easy to permanently damage.
It's the epitome of fitting a solution to a problem. Cooking and cleaning are simple tasks that any able bodied person can do themselves. There is no shortage of able bodied people in the world.
In this case, clothes washing machines and vacuum cleaners seem to fit your thesis.
On a tangent: How was wall-to-wall carpeting cleaned before the invention of the Hoover? Did it just not exist before then? Or it just collected dirt until it was replaced?
Still useful for collecting crumbs around restaurant tables.
Washing machines freeing enormous amount of time for people (particularly mothers, who could use that time to e.g. read books to children or educate themselves) are one example. The other I like is the Reformation: why did Martin Luther succeed where previous attempts at reforming the Church have failed? Because the newly invented printing press allowed for the theses to spread across the Old World much faster than the Church could control it.
Nuclear power was an existential threat to coal mining but was derailed almost completely by manipulating people's beliefs.
The economist Dierdre McCloskey, for instance, argues forcefully that the revolution started with the idea that all souls weigh the same -- i.e., people are all equal under God. [1]
That conviction led to the belief that people should be allowed to read the Bible for themselves, which led to widespread literacy, which led to a demand for books, which allowed the printing press to survive and spread.
After much to-ing and fro-ing I think I'm more inclined to McCloskey's view, that cultural beliefs and values determine institutions, and so determine which technologies survive and which die out.
Of course these things are not one-way. Technology affects beliefs and institutions. But other things do, too, and McCloskey thinks the other things are at least equally as important.
[1] McCloskey D., Bourgeois Equality.
I'd gladly give up houshold chores if it was cheap and wouldn't require a person coming in my home.
I’d buy a robot though.
House chores and maintenance are two big things that are yet to be tackled by technology. They're quite literally life wasters - things you must do in order to get the things you want to do. Prime candidates for automating away, the opposite of "fitting a solution to a problem".
Where I live everyone has a robotic robotic lawn mower.
But cutting out laundry, dishwashing and lawn mowing still leaves some time sinks. Like cooking, where machines currently solve the ingredient prep and the actual cooking parts, but do not help much with cleaning up the mess afterwards (beyond the baseline of using a dishwasher). Or dusting, laundry folding, general cleaning. These tasks I'd wish they could disappear.
And then there's the general category of maintenance - most of the products around us require regular care. Which gets annoying rather quickly.
The tech we have is cool, but there's so much more to be done! I'm hoping the future will eventually bring self-cleaning and self-repairing materials.
Wait... Isn't fitting a solution to a problem exactly the right way to do things?
For instance, if you were replacing a human salary, it might make sense to purchase a hundred-thousand-dollar robot, but if you just want something to pull weeds in your yard once in awhile, it might be hard to justify spending more than a couple thousand. If the parts alone cost tens of thousands, then that's a pretty big economic limitation on how the technology gets used.
[1] https://www.bostondynamics.com/spot
AI vs AGI is well understood in cognitive tasks but tasks that require motion would also require a parallel breakthrough.
If "lifetime" can be considered to be the next 50 years, I absolutely disagree with you. I'd even venture to say that we will see robots with human level walking skills within the next 15-20 years, at most.
I remember looking at some original robots developed around midcentury in England. They were inexpensive and very bad but did function basic functions.
I look at the Boston Dynamics robots and see these are very expensive robots performing more natural movements.
Let us suppose the BD and similar teams are at the half way point.
To get human level performance at this rate of change means trillions of dollars. This robot would impress us - just as the original English robots and BD robots do. The French Empire invented expensive and difficult to create toys and no robot revolution started. I don't see anything continuous between the idea of a robot and the idea of machine automation - I think that is too much generosity. I know people want to give it because the field is very difficult - but this is not being honest.
The ideal of the robot - people have different ideas about this - but the ideal for me is that there is an economical and effective alternative to human level physical articulation. I'm sure you have more advanced definitions - this is just the basic benchmark. It should be plausible for a robot to sew a thread or squeegee a window with grace.
It is not just human level articulation - it's also generalist application.
General purpose machine articulation looks like a high bar to the point it is difficult to imagine what the world would look like were we to be successful.
I recommend you check the work of the other Simone out - it is delightful.
Because of stairs, etc.
But maybe it would
1. allow them to stand
2. allow them to use stairs
I just asked if we could strap people to these things to let them climb stairs, lol
I think unfortunately the main reason the iBot did not originally succeed is that it was too expensive.
And, if anything went wrong halfway up the stairs, you'd be going down ass-over-teakettle strapped to something metal weighing more than you.
All those things could be improved on. Maybe Mobius will succeed.
Right now when it's time to carry that heavy desk, couch, piano, etc. through doorways, halls, up/down stairs, etc., two really strong people do it manually.
Imagine if they showed up with a robot that could do it. In addition to walking, it would need to understand gripping and balancing heavy objects and how to orient and maneuver 3D objects through tight 3D spaces, but I bet those things are understood well enough.
Unusual physical strength, or powered mechanical aids, are mostly only used with unusual items. Ex, most "gunsafes" are more properly Residential Security Containers, and aren't all that heavy (~200lbs) and can be moved around, including up stairs, by two average adults. Real safes weigh more like 600lbs and are much more rare, partly due to the difficulty and expense involved in moving them anywhere remotely tricky. I have heard of powered stair climbers being used to lift such things up a flight of stairs.
[#] Mountain-Boting is now a word