The video discusses the limitations current robots face due to their reliance on outdated industrial technology, which makes them loud, rigid, and lacking adaptability. Most robots are only capable of performing one repetitive action safely at a time and are often designed to shut down when humans come close, due to safety concerns. It contrasts this with advancements in AI, like ChatGPT, that can virtually interact but highlights the lack of equivalent progress in robots that can flexibly assist in daily activities with humans.
To overcome these limitations, the speaker introduces an innovative approach that takes inspiration from nature to create a new type of soft robot. By using a combination of soft and rigid materials to create musculoskeletal robots, similar to biological structures such as human muscles, tendons, and bones, this technology allows robots to achieve versatile motion skills, previously impossible. Practical examples include a robotic hand and leg that mimic natural human and animal movement using these new technologies.
Main takeaways from the video:
Please remember to turn on the CC button to view the subtitles.
Key Vocabularies and Common Phrases:
1. versatile [ˈvɜːrsətl] - (adjective) - Able to adapt or be adapted to many different functions or activities. - Synonyms: (adaptable, flexible, multipurpose)
But how can we translate that expressiveness on paper into versatile motion skills for robots?
2. rigid [ˈrɪdʒɪd] - (adjective) - Unable to bend or be forced out of shape; not flexible. - Synonyms: (inflexible, stiff, unyielding)
The electric motors and their links, they're made of metals, which makes them just inherently loud, rigid, and also a bit unsafe.
3. musculoskeletal [ˌmʌskjuləˈskɛlətəl] - (adjective) - Referring to the muscular and skeletal systems, involving both bones and muscles. - Synonyms: (muscle-skeleture system, body structure)
We use this musculoskeletal design to create a robotic hand with many densely packed degrees of freedom
4. teleoperation [ˈtɛlioʊ-ˌɑːpəˈreɪʃən] - (noun) - The operation of a machine at a distance. - Synonyms: (remote operation, distant control, telecontrol)
What's great about this teleoperation is actually we can teach the hand new motion skills just by telling him what to do.
5. adaptive [əˈdæptɪv] - (adjective) - Able to adjust to new conditions or environments. - Synonyms: (adjustable, modifying, resilient)
This like learning like a child allows us to actually become a bit more adaptive and also a bit more robust to unseen scenarios.
6. electrohydraulic [ɪˌlektroʊhaɪˈdrɔːlɪk] - (adjective) - Relating to the combination of electrical and hydraulic systems. - Synonyms: (electromechanical, hydraulic-powered)
What's great about this electrohydraulic muscle is that it not only is actually quite fast, quiet, and more efficient, it actually also contracts like a real muscle.
7. antagonistic [ænˌtæɡəˈnɪstɪk] - (adjective) - Working against each other, often describing muscles that work in opposition to accomplish movement. - Synonyms: (opposing, contrary, counteractive)
This leg jumps by using antagonistic pairs of muscles.
8. composite [ˈkɒmpəzɪt] - (adjective) - Made up of various parts or elements. - Synonyms: (compound, complex, blended)
And that allows us to use things such as lightweight composite fibers to make the skeleton.
9. ligaments [ˈlɪɡəmənts] - (noun) - Tough, fibrous bands of tissue that connect bones to other bones in the body. - Synonyms: (tendons, sinews, connectors)
We use durable string to make the ligaments and the tendons, and also soft silicone rubbers for the skin.
10. revolutionize [ˌrɛvəˈluːʃəˌnaɪz] - (verb) - To change something radically or fundamentally. - Synonyms: (transform, overhaul, modernize)
And while this technology will hopefully revolutionize healthcare and space exploration, it will soon, along the way, allow me to get my folds, my clothes folded, my kitchen cleaned.
How we can make robots more human-like - Robert Katzschmann - TEDxGateway
What comes to your mind if you think of a robot? Well, maybe this metal machine assembling a car in a factory, or this round object going around your home and vacuuming a floor. Some of you might even think of a robotic dog that is inspecting a plant. But have you ever interacted with one of these legged creatures in your daily life?
Probably not, because the thing is, you know, most robots, they actually do just one rather repetitive action, and they do that well, as long as we humans don't come in and interfere. Did you know, actually that most robots have to shut down once a human comes close because we might get hurt by them? So in order to make robots that are truly useful for our daily life, we have to think of something that they can do much more than just one thing. And they should do that by smoothly and gently interacting with us. So let me ask you, what would you like a robot to do for you?
I see you thinking, okay, okay. While you're at it, I tell you I would like a robot to just fold my clothes or clean the kitchen. But that's not possible yet. Well, we can already ask ChatGPT almost anything and will come back with an extremely impressive answer. We just saw that. Right? But how can we translate that expressiveness on paper into versatile motion skills for robots?
Well, I think these versatile robots are actually not too far off yet. As long as we start to create robots that are actually gentle and safely integrate into our world. Most robots these days, even the most advanced ones, they actually still are based on the technology of, of last century's industrial machinery. Their electric motors and their links, they're made of metals, which makes them just inherently loud, rigid, and also a bit unsafe. So why don't we just stop building metal machines and instead take inspiration from nature to create softer robots that are actually integrating into our natural world?
That's what we've done when we created Sophie. Sophie is a soft robotic fish. She's right now swimming in the ocean, as you see here, and inspecting marine life. We actually believe that robots should not be as rigid as we have them today, but they should be more squishy like a fish or cuddly like a dog. So the rigid bones in our body are actually much softer than these metals in these machines. And while we need the bones to function, to move in some way, truly agile humans can only be agile due to their muscles, to their tendons and to their ligaments, and these are even softer than what the bones are.
Imagine you waking up one morning and everything that previously felt soft has suddenly Become as rigid as metal. Wouldn't it be impossible to hug anyone? So that's why we create muscle skeleton robots made from a range of soft and rigid materials. And that allows us to use things such as lightweight composite fibers to make the skeleton. We use durable string to make the ligaments and the tendons, and also soft silicone rubbers for the skin. We use this musculoskeletal design to create a robotic hand with many densely packed degrees of freedom. We have to do this because otherwise we will not have ever a chance to become as nimble and as versatile as a human hand when it's interacting with an object.
So let me ask you, shall I ask Benedict fore our machine learning scientists in the group come on stage and fire up this thing. All right, so what you see happening now is our robotic hand starts to move its individual digits. And yeah, that's fine, you can clap if you want. We have individual bones here. We're connecting them with tendons, ligaments, and a rolling contact. So each individual element that you see here was designed, produced, and then assembled in our lab in order to even have a chance to mimic nature in the way it moves.
So what I'm going to show you now is here we have a camera. This camera allows me to take my hand and I can move my fingers and the hand will just do exactly what I'm doing. What's great about this teleoperation is actually we can teach the hand new motion skills just by telling him what to do. I just move my hand and it does everything. But the thing is, I cannot always babysit this hand every step of the way. Wouldn't it be much better if I can just give this hand a high level command and it goes ahead and figures out everything else by itself?
Wouldn't that be great? That's why we develop algorithms that allow the hand to manipulate objects on its own. I brought you here a cricket ball. Would you like me to try give it to the hand and see what it does? Yes, sure. All right. Okay. So as the hand is getting ready, it's probably still thinking a bit. It will try to use its fingers to rotate the ball in its hand. So what you're actually seeing here was not pre recorded. This motion was learned from scratch. We use many paralyzed simulations to condense months of training into just hours. This like learning like a child allows us to actually become a bit more adaptive and also a bit more robust to unseen scenarios.
And you might wonder, we cannot just only manipulate a cricket ball. We can also pick up tools that we humans would usually use. They are more complex in their nature, and our hand can then just use its thumb and start operating it. So before I forget, let me get back my ball. Thank you. And shall we see if this robot is actually polite? Oh, it can shake hands.
Look at this. But wait a second. So before I let go, take a look at its wrist. You probably can't see it, but we still use noisy and inefficient motors, and we have to replace those with artificial muscles. I've told you about them earlier. So, yeah, I think you like me, but you can let go. It's okay. Thank you.
So would you like to see how these artificial muscles work? Okay. We take pouches, we fill them with oil, and we cover them with electrodes. We then apply a voltage onto these electrodes which causes the liquid that's inside of this pouch to be squeezed to one side. The whole thing then contracts. What's great about this electrohydraulic muscle is that it not only is actually quite fast, quiet, and more efficient, it actually also contracts like a real muscle.
So to give you an idea of how, we have now created a robotic leg using this technology, and this is the first time that this is being shown. We made four packs of these muscles put together and brought them to you today. Shall I ask Thomas, our robotic wrangler, to come on stage and fire this demo up? All right, so what we do here is this leg jumps by using antagonistic pairs of muscles. One side of the muscles contracts while the other side expands. This makes the limbs of the leg move and jump.
A rigid robotic leg could do this, but I can tell you it would get stuck in such a setup. We have used the adaptiveness of the musculoskeletal design to go over rough terrain and just different terrains using the same approach. And I can tell you we are just at the start of the journey. Can you imagine how much better these robots will become once we even start to optimize their design?
So I believe that robots of the future should smoothly and safely integrate with our lives. This musculoskeletal technology that we've developed allows us to create robots that use ligaments, tendons, and muscles, and they will get us there. We have to then look forward to robots that are going to be more adaptive, more efficient, and more quiet than what we have today.
And while this technology will hopefully revolutionize healthcare and space exploration, it will soon, along the way, allow me to get my folds, my clothes folded, my kitchen cleaned, and just focus on more enjoyable things.
MOTIVATION, INSPIRATION, TECHNOLOGY, SOFT ROBOTICS, MUSCULOSKELETAL ROBOTS, INNOVATION, TEDX TALKS