We have two points of contact, on a good day. Our legs are pendula. Our ambulatory motion is essentially a constant form of falling forward. We rely on accelerometers inside our auditory canals. We shift our limbs, our superstructures, our very organs just to stay upright. We're a goddamn mess of an animal.
@@blink182bfsftw We know Africa went through a drying event that caused savanna formation. General thought is the shrinkage of forest habitat forced us into the savanna, while our cousins, the chimpanzees and bonobos stayed in the forest. In the savanna the ability to stand and see over the grasses was a huge advantage. Many animals stand on their hind legs to see over things, dogs, squirrels, bears, mink, muskrats, most omnivores and carnivores have some capacity to stand upright. The difference with us is that we were already part way to upright walking. Our cousins can do this, for short distances, though they don’t walk well. If our last common ancestor was fairly similar to the physical layout of chimpanzees, the switch to upright movement wouldn’t have been as big as it would be for canids or felids. Another reason is freeing up hands. Our cousins knuckle walk, which makes carrying things difficult. Babies for example. I personally suspect it wasn’t a single factor, but rather a confluence of several. But I’m a student, and that’s a gut feeling based on experience from another field.
Humans aren't designed perfectly, but we need to develop robots to walk around our buildings. It's like QWERTY: it's not optimal, but it's the standard.
Yes, apparently QWERTY was developed for mechanical typewriters to deliberately slow down human typists. Apparently they were fast enough to jam the mechanism on a regular basis. Currently tools and buildings are designed for humans, and that won't likely change in the near future.
Moravec's paradox highlights the inherent challenges robots face in performing tasks that come naturally to humans, due to their limited sensory and flexibility capabilities. In essence, we are asking robots to carry a plate with only a hammer at their disposal. To enable more adaptable robots, it is crucial to invest in enhancing their sensory input and flexibility, allowing them to better mimic human complexity and versatility.
This is exactly why Tesla is trying to conquer this next. If you can create a humanoid robot, all of the Technologies you would develop along the way will be incredible. Tesla wants to be a premier robotics company and if you can create a humanoid robot then you basically can create any type of robot, if you want
@𓆏 doubting Nvidia is dumb. Really really dumb. So, since Nvidia is the one developing the solution... Ya, it's gonna get solved. Nvidia generally beats my expectations or releases things I wasn't even thinking of or was thinking was possible. They probably also enable skynet and get us all killed but well, win some lose some....
@𓆏 I don't think is an algorithm issue even if now is both an algorithm and a hardware problem, let's say tomorrow we get the algorithm that allows the robot to perfor actions in the real world, you ask him to clean the dishes, imagine how much data come to your brain when you perform a task like that, how many things your fingers can feel and how incredibly accurate the movements are, you fingers can feel, temperature, pressure, grip, wetness among probably many others, and can perform movements that are very precise, we can't replicate that. To drive a car I think will be done far sooner than we will have robots capable of doing fine delicate tasks, I believe will have to be done biomechanically, we can't even dream of building anything even close to the complexity if a human hand with all its capabilities.
True, but consider the sheer computational power of the human mind. From what I’ve read, only the largest supercomputers on the planet are able to match it in calculations per second. Unless computation gets much faster, lighter and cheaper we likely won’t be able to have an average machine running on the same quantity of information the human mind processes. If that’s the case, we will have to figure out how to make the hammer perform brain surgery, which should be possible albeit extremely difficult.
Trivia for all here: The word "robot" was first coined in the Czech play "Rossum's Universal Robots" by Karel Čapek which premiered on 2 January 1921. In the play, the "robots" are biological rather than mechanical (kind of like Replicants from the Blade Runner movie).
Fun fact: The word itself derives from the Czech word “robota,” or forced labor, as done by serfs. Its Slavic linguistic root, “rab,” means “slave.” Therefore, robot is a slur for robots.
I propose a robotics version of the Turing test, which I call the Pooing test. To past the Pooing test a robot must be able to change the nappy on a human without traumatising them or getting a mess all over the place, then they would have proven that they can replace humans in the lowest paid menial work.
I'm not sure that's the most menial work robots do, but it would be a critical milestone never the less. The most menial and dangerous work some robots do includes sewer systems inspection like the Redzone Solo, working within the bowels of the Chernobyl and Fukishima Daiichi Power Plants, and performing full GI endoscopies like Ryukoku University's camera pill robot dubbed the 'Mermaid.'
@@PaigeWylderOwO I don't know if menial is the right word. Changing a nappy is a REALLY hard problem for a robot. So is folding laundry. These things are non-trivial for humans, even.
Post spine surgery, in the ICU for 3 weeks, bless the hospital staff for changing my nappies four or five times during day & night... I'll never forget them for their gracious and kind care.
@@0neIntangible Yes, it is no joke, a very important task, not just for infants either. Imagine having that sort of robotic help at home if you are a quadraplegic. It would be life changing.
Soft robots are a research toy. Humans are somewhere in between rigid and soft. A hand is basically like a rubber glove but with calcium rods and plates embedded in it to give it rigidity. Also, if you make a force actuator weak and soft (compliant) you can also use it for feedback detection, like a spring scale with an adjustable strength spring.
That only concerning grip and movement but a human hand is far more complex that that, just your fingertips can give your brain all sorts of different data from pressure, to temperature, grip level, texture, shape etc. You need all that information if you want to manipulate the world around you and I don't think what we see today ( hands grabbing something) is even close to good enough to be useful
@@dumyjobby Indeed, I am only providing a brief description in a youtube comment. Simulation of skin touch would likely utilize tiny piezoelectric stress elements to detect distortion of an otherwise flat silicone "skin" sheet, and tiny thermocouples to detect temperature. Thermocouples have a hot and cold side, and our bodies work similar. We generally only detect heat and cold relative to our inner core temperature. The body can be fooled by putting one hand in very warm water and the other in ice water for a few minutes. Then place both hands in lukewarm water. The hand that was hot will feel cold, and the hand that was cold will feel hot.
For some reason, I was thinking there would be more depth in this video considering the vast history of humanoid robots, but I learned about obstacles I didn't know about in locomotion like Zero Moment Points, and appreciate this video never the less. For future essays offering further details on humanoid robotics throughout history, you might want to look at the DARPA Robotics Challenge between 2012-2015, the Nasa Robonauts, and in the indie development sphere, the MaSiRo Project, among others.
@@brodriguez11000 "ChatGPT, write an essay on the development and challenges of robotics and make it understandable for low-knowledge youtube crowd". There.
Nice intro with Astro boy. Arguably the first popular version of this story is Pinocchio. Then again earlier still by a few millennia is the Greek myth Pygmalion. An artist creates a beautiful sculpture of a woman, brings it to life, falls in love with it and being a Greek myth, things go badly, and tragedy ensues.
its fascinating how the concept of a synthetic companion goes back so far in history. Even in some ancient mythologies Gods would create Golems, Titans o Giants from inhert materials to serve a purpose.
As a person in robotics research, gotta say a well made short summary of where we're at! It's still a open question if some one the method you described at the end will work for humanoid robots by just "scaling it up" like we've seen for Deep Learning. Simulation training and using that (Sim2Real) have been around for about 8 years now since Domain-randomization seemed to show some promise, but so far have had limited success for task that is inherently complex to simulate. It sure is an interesting time. Yann LeCun (Head of Research at Meta) claimed that in the future we can make a robot work in the real-world by just training it on youtube videos and create a "world model" from the physics it sees. Slightly skeptical if vision-only will lead to this capacity but if it does, more generalist robots might happen in our lifetime.
Very nice overview of the field and its evolution, but why no mention at all about an important new player Tesla’s Optimus. It would also have been very interesting to find the reason behind Honda stopping the Asimo project. They were not far from succeeding and their platform with the recent AI developments would have made it a very strong contender to the house assistant job.
I thought omitting Optimus was an odd choice, as Tesla has already major strides (heh) in its development, including demonstrating working around humans and the soft touch that he mentions being a necessity. Still, there are SO many players in the field right now that it would be impossible to name them all.
Simulations have come a long way, so I imagine the field of robotics will be moving quite fast. Better aligned language models should help train and make robots safe around people... I sure hope these come out soon. There's housework to be done.
My personal benchmark on when we will have solved Moravec's paradox is when we can build a successful "rummage robot". Such a robot could, when presented with a pocket full of small items of varying shapes, sizes, and textures (keys, wallet, cards, gumstick, earbuds, etc...), reach into the pocket, rummage around, and pull out a specific item while leaving the other objects in place. Such an act, simple for a human, requires a massive combination of different capabilities that robots are currently many years/decades away from being able to accomplish.
@@MegaChickenPunch I'm talking a realistic pocket like what humans have on their clothing; pressed against the body and full of junk. A camera isn't going to do you much good jammed into the pocket and pushed up against the items, not to mention that doesn't solve any of the grabbing and actuation side of things.
12:53 This robot is now better than a toddler. Go ahead and tell me how many toddlers will pick up and put away all their toys, especially the first time you ask?
The really scary perspective is when you consider that humanoid robots will inevitably eventually be "enrolled" in armies. Initially in order to avoid human casualties. The danger is that there would be little political penalty for starting wars and that could make them more frequent. Countries would race to spend precious resources on "divisions" of robots in order to avoid being outnumbered by their preferred rival countries. Imagine US vs China with industries churning robots like... there's no tomorrow. Unmanned weapons are already there - and they have no issues collaborating with each other (contrary to real soldiers on the operations theater).
Even if we somehow can simplify our manufacturing process to built mass manufacturing factories for "war robot" construction. The amount of valuable resources needed to built a humanoid AI in such a large scale is enormous. Especially when China is sitting atop the world largest rare earth deposit. Furthermore, a human soldier may takes a long time to grow up, is weakness to physical limitation and fear of death but most of the cost to maintain them are foods and pay ( and that is assume if they manage to survive the war or survive till the end of their contract ), so you can argue that in the end, the total cost to train and equip a human soldier is much less than a "robot". While with the case of sophisticated advance robot, the military / government will have to pay a large amount of money up front to purchase one. And imo, there will still be political penalty and casualties, especially when it comes to civilian casualties. Powerful weapons are not know for their ability to limit the amount of collateral damage caused by them
Not armies. Police. When you don't need people to force other people to obey your rules, you don't need to give said people "rights". Machines could just hunt down and destroy any electrical signature, and whole populations could be relegated to stone age existence, and those who control the machines could just forget about them, as they would never be able to post any threat. Like the millions of people trapped in the Gaza strip, machine gunned down any time they try to escape... but without requiring people to do the shooting, there's no capacity for conscience to creep in, no whistle-blowers, and eventually, only myths told about the "dangers" kept inside there by the machines. We don't have a good history for how we treat those we don't see any need for. People not even worth enslaving.
@@favesongslist yeah, it should be mention in video about word robot: Robot is drawn from an old Church Slavonic word, robota, for “servitude,” “forced labor” or “drudgery.”
Astroboy... There's been Metropolis (Fritz Lang) before. And the word "robot" comes from a novel by Karel Čapek, RUR (Rossum's Universal Robots, 1920). The Golem, from the Hebraic tradition is somehow a humanoid robot. We've had many humanoid automata during the 18th century in Europe too. So, indeed, the Japanese have become very inventive and enthusiast with the robot companions, but the concept comes from the Western world.
14:54 The code in the example actually makes a big mistakes. The instruction is move "while keeping 12m away", but the code ONLY sets the FINAL target position to 12m away, which would totally allow to pass much closer to the turbine while in transit in certain spatial arrangements, even leading to a crash.
I haven't thought about this recently but I think AI training will do a lot to help robots mature in their complexity of movements. Just show the training super computer a bunch of videos of humans doing a task and they will figure it out. Then you can install the optimized program in all robots and its basically a solved problem. Will be much faster than a kid learning it eventually. Thats where things get really scary. In the matrix they could download programs in their brain to learn things but IRL it will be robots that get this ability, not us.
Quite possible, though your training method is a bit off. Wrong sort of data, too much noise. A better way would be to first write a program to simulate a body - that's a fairly simple task, basic rigid-body physics model. Used all the time in computer games. Now you have a source of limitless training data, which can be easily adjusted to your particular robot's body shape and dimensions.
My best examples i give people when they say, more automation is, I ask them, what would it take to lift, open and dump a 20kg bag of flour into an industrial mixer. Gentle. Meaning you first drop the so it covers the surface, wait for the mixer to mix that flour then repeat until the flour bag is empty. Needs to be able to handle a flour bag that gradually gets lighter, it needs to know how to stop, it needs to never drop the bag into the mixer, it needs to do so many different things. So it is cheaper and easier to simple hire a man to do it.
Why not use a claw-machine style grabbing, holding, and moving motions with suction/rubber technology maintaining the 'hand' balance? Instead of soft or rigid grabbing?
I find it amusing that the current spate of news about AI almost always uses images of robots, even though the two are not the same. Certainly we hope to have an AI that will run a robot (i. e. Tesla's Optimus), but currently the most mechanically advanced robots are not autonomous at all, so why conflate the images?
Why do we need humanoid form as opposed to ideal functional form? The automobile looked nothing like the horse and yet everyone got used to it. We would be better off letting form follow function.
I love your work, there's something very important about being curious of the world. Most ppl I meet day to day aren't bothered and that's a shame Indeed.
* QUESTION * Assuming all the scientific and technical challenges are solved and human-level intelligence is achieved, based on a reasonable estimate of what robots will cost, what percentage of jobs will be replaced by robots? Answering this question might give us a glimpse of what the world will be like in ten years.
This is why India is lagging behind the rest of the world in humanoid robotics. Can you imagine the mechanical complexity in recreating all that head motion!
excellent video . Before it came up at the end of the video, I was thinking "Why not make a 4 legged robot with some sort of arms on top ?" Looks like Boston Dynamics are already doing something like that ! I guess humans wouldn't mind in their homes a robot that is like a dog with arms on it's body, or maybe like the Greek mythological creature of Centaur, half horse, half human. The feet could have rubber tyred wheels on them, why bother with all the leg movements when a wheel would be better in some situations. The wheels could be braked for when leg movements are needed to overcome an obstacle. I guess someone has already thought of that too !
I feel like with time humanoid robots will help us out with projects like: Buildings Factory working And even exploration. I just hope things don’t turn out like Detroit become human
The solution is not using a robot. It's like trying to solve every problem with a hammer. We have to modify another animal and create new one to set the table
It really sounds like a new generation of analog computers are gonna be needed. You could get all the computing power you need in a small and energy efficient package.
They have things like Isaac Gym to train machine learning models for robotics control within a physics simulated virtual environment. This scales like crazy.
What is a humanoid robot? Did you mean android? Oh whoops, that context has already been used for some bizarre reason. (Our phones are not androids. Humanoid robots are androids...)
Why not just print biological muscles and biological limbs and nerves? This would also allow for testing of biological to electronic interfaces. It seems that in trying to develop systems that mimic biology in a completely mechanical system is a somewhat backwards approach of trying to reinvent the wheel
In my opinion simply mimicking biological systems will not really help us understand how these work. We didn't invent airplanes by simply flapping artificial wings as birds would. We discovered aerodynamics which in turn helped us understand how birds fly. We can now fly faster and farther than any bird can.Printing muscles, limbs, nerves and such will also not help us build humanoid robots. Might as well just birth a human. Certainly more easier and relatively more cost effective.
@@AndraBullar Yeah. Makes me wonder if natural movement really is a product of a computational process. I think it's a system that is governed by reactionary processes meaning that biological components are merely reacting to external stimuli. Just standing still is an interplay of organs that give feedback to each other to keep us balanced and upright.
I believe the name of the game in complex and adaptable robotic systems is efficiency. Currently biological structures beat out man made structures in terms of compactness and energy efficiency of movement
@@rockprime1136 We can fly faster but not necessarily better, wings are quite a bit more efficient and far more agile. Have you seen a bird fly through canopies or a dragonfly catch a fly midair in slow motion?
Biting more than you can chew is the hallmark of idiocy - why make a robot that can walk, and is limited to only a couple of 3-jointed-hands (the 3 joints being the shoulder, elbow, and wrist) if the task can be easily accomplished without it? Just bolt several (lets try 5) multiple-jointed (5 joints - like a hand with 2 extra elbows for added mobility and reach) hands to the ceiling, and install cameras to see all the kitchen without blind spots, add a trained AI, and viola - a robot that can set and clean a table, wash the dishes, and cook like the best chefs in the world. Then, mount these "hands" to a network of rails on the ceiling so that they will be able to travel through all of the house to get the next generation of Roomba... It is just like with hydroponic farming - why keep the plants resistant to excess sunlight, rains, pests, and diseases, if all these problems are 100% solved by the plants growing inside a warehouse?
Dream is right. Who needs a robot that looks like a human? To drive a an old pre computer car for you? What good is the human shape? Robots have been in service for decades doing all sorts of tasks. There isn't one that really makes sense to look like a human. What I need to interact with a fake plastic smile?
Speaking as someone who suddenly woke up one morning with major arthritis in a big toe joint..... You really have no idea how important the toe push off is to walking. Seriously.
09:40 Rookie mistake! I spit into my hand to wet it, place the flat palm onto the plate followed by creating a suction cap effect with it in order to lift the plate off the table! Checkmate!
Let me clarify for OP, you (read: the worker) are the robot. The machines simply make it so that you may robot more efficiently (read: make your employer more money) An alternative way of putting it, is general AI and a fully automated workforce is just the dream of the ones who believe that their doomsday bunker will somehow staff and repair itself.
I believe the biggest challenge will be in making anything scaled down to a human size. And you know what that means? We're getting a real Gundam sooner than later!
At the end of the day the feasibility of most these emerging technologies comes down to batteries. As soon as we have energy densities that are reliable, efficient and safe enough the sky is the limit.
the moment you figure out how to create intelligence (which already happened), the advancements in intelligence will always be more valueable than any other advancement, because it will always outpace it and quickly replace established technologies. we will see this happen over the next few months. people will develop all sorts of sophisticated soft and hardware just to be obsolete half a year later. this will be a clear sign of the singularity happening.
I'm pretty sure robots will evolve the same way AI has evolved: almost nothing for fifty years, and then some mathematicians will publish a paper called "All You Need Is Sense" and soon we will be inundated with humanoid robots in a matter of weeks.
Humanoid has been a pretty well defined term long before robotics was even a pipe dream. Anthropology and science fiction use it to refer to anything human-like in shape.
The insane amount of technical complexity and potential failure modes of bipedal robots make either 4 legged or wheeled robots a lot more attractive. Sure, a bipedal robot could, in theory, do everything a human could do. But do we really need a robot worker to be able to do everything a human can do? Is it worth a huge price increase just to make a robot that will spend most of its time moving over flat ground, or standing in place, able to climb a ladder? There's a reason why most robots that do serious work are 6dof arms bolted to the floor.
I'm skeptical of the need for humanoid robots. Yes, it would be cool, but I think people would prefer interacting with human personnel rather than with machines. Also the tasks that are most likely to be performed by it are performed by low paid workers so it wouldn't be really cost-effective. Maybe there would be a potential for robots in rescue operations or in space explorations, but still humanoid design probably isn't optimal. Great video as always!
So the robot will be telling the humans of the future " that is impossible you cannot walk on a tight rope across a waterfall ". Then we do it. Great update the walking section was fun that's gotta be tough goal.