Strength is additive. Why can't you just add more smaller muscle fibers that work in concert? That's basically how real muscles work. This seems so obvious there must be something wrong with the idea.
Great question. And yes, you can have lots of smaller ones to do the work of a few larger ones. The strength (stress) comes essentially from the cross sectional area of the muscle (or bundle of). And no matter how thin you make the spaghetti, you still need an equivalent volume of it to get the same cross section, For a given length.
The reason speed, as in how many times a second you can make it contract and lengthen, is a limiting factor comes down to good old force = mass x acceleration. Want to make it go faster?then you need higher pressure. So now You have a lot of water you need to move at high pressure with very rapid acceleration. This means the scale of the pumping equipment gets prohibitive very rapidly.
Real muscles work at a fundamental level by a kind of ratcheting mechanism between proteins that pull each other together. This process can happen very quickly. Mammal muscles don’t need to pump lots of liquids around with high acceleration and velocity. So there is actually a lot less inertia from moving mass in real muscle.
Another issue I didn’t touch on in my earlier post is that this kind of muscle is very difficult to do strain and force sensing with. To make motion with this kind of actuator that is both rapid and smooth is extremely difficult and still an open problem.
But the problem remains that there is still a severe trade off between speed, force and backdriving. The Spot really doesn’t have that high of a max force it can generate, and no where near that of a similarly sized dog. Additionally, the motion produced by the kinematics system it uses it relatively simple with far fewer degrees of freedom than a real dog.
The general consensus in the field is that if you want to have a robot which can do everything a human can do, you need to at least match the degrees of freedom a human has in motion. It would be very difficult to fit 100+ harmonic drive servos into a robot. Of course controlling all of this effectively is extremely difficult and that’s often why robots like spot use vastly simplified kinematic models, because it makes the equations go from near impossible to solve to just hard.
series elastic actuators are promising and I have high hopes for their use in the future. I have some colleagues who work with them. The limitations there are more to do with complexity and finding ways to dynamically tune the force impedance of the springs.
I can say that I’m a big fan of K. Eric Drexlers work and that you should read it if you want to know what the future will look like.
I suspect the Festo muscles will end up in mostly industrial products which need simple, linear motion, and weight and efficiency don’t matter as much. Eg manufacturing equipment and active vibration damping systems.
Would love to be proved wrong on this however.
On the criticism, do you think more limited / lab-bound systems will help develop control technology in advance of better muscles?
Long answer: I think that the more minds there are working on this (or any other interesting) problem, the more progress will be made. And we shouldn’t let the perfect be the enemy of the good enough.
The field of robotics, and especially soft robotics like this, is so new and uncharted that individuals tinkering in their garage can and have made real advances for the field.
That said, really good control systems for these kinds of actuators are something very much lacking. The human nervous system, and in particular the vestibular system is so absurdly effective, efficient and complex all at the same time that it boggles my mind. Nothing anyone has ever designed has ever come close to that level of functionality. But to make a robot that can do human things, we need that level of functionality. We need to have really good proprioception which is integrated with force and feedback sensing.
To give an example; open a bag of potato chips and pull out a decent handful. Now eat the chips one by one out of your hand without dropping any and without using your other hand. I dare say you can probably do this, and without crushing the chips in your hand. You can feel just how much force to apply, you can feel where to place your fingers so that all the chips are held just so. You can make subtle adjustments after each chip is taken away.
There is no robot hand in the world that can do that simple task. Ask any arm amputee. And to get to that point, a lot of really difficult stuff has to be invented and worked out. I suspect this is a problem which will be solved piecemeal, over time, but much of the fundamental work in things like the mathematics of motion control, material science and compliant mechanisms will come from very well funded labs. If I had to guess, I give a 60% chance a robot can do the chip eating task within a 5 years. And 85% within 10 years. (Especially if the work of my team pays off)
Is the technology dealing with interfacing with the brain good enough, even if on the arm side it's not?
I work in robotics, specifically compliant and modular actuators, similar to the ones in the video.
As I see it, (and many others in the field) One of the biggest issues in robotics is a lack of good artificial muscles. The need for them is that in a robot which can work with people you want it to have both high enough force output and power density to actually move things, but you also want the robots arms to be back drivable; you want to be able to move its arms by an outside force. For welding robots you want the opposite, zero backdriving and zero backlash.
To get back-drivability, good torque and reasonable power density, you generally can’t use spinning electric motors. You need to hear them too much. This is why Boston dynamics use hydraulics. It’s also why Boston dynamics is unlikely to ever sell you an Atlas in its current form because those custom hydraulic actuators make the system so unreliable and expensive, even a mining or gas company would struggle to justify it.
The short and long of it is this: if you want a robot that can do the things humans can do, like play soccer and rock climbing and knitting, then you need muscles like humans have.
So you need artificial muscles. There are a lot of different ways of making them. [1] the issue is no published technology is able to give you want you need to make a good analogue of human muscles.
The muscles in the video are (correct me if I’m wrong) soft fluidic actuators. These actually have pretty good potential, especially compared to things like piezoelectric actuators. But when taking into account how you would actually put these in a real world robot, problems arise.
But the main downside is you need lots of pumps and tubing to make it work, and it has limited maximum stress and relatively low actuation bandwidth. In a nut shell, to make these strong enough to do much useful work, you need them to be really big, and that means they can’t move very fast. There is a reason you don’t see any commercially available robots with this kind of actuator.
I really hope this creator can make his system do what he has set out to achieve. And perhaps he will be the one to figure out a new kind of actuator which does tick all the right boxes.
Conflict of interest: The startup I work at is developing a commercial product using a new kind of artificial muscle. So my views may be coloured by that.
[1] See this paper for a good breakdown of artificial muscles: https://sci-hub.se/downloads/2019-10-12/7c/zhang2019.pdf?ran...