I'll admit that I'm not a scholar on the matter, so this may just be ignorance speaking ... but I wonder why all the soft robotics stuff that I see never seems to pair the soft/flexing actuators with any rigid components?
I'm thinking of the analog to the human body (or animals) ... we have a skeleton, organized and connected via ligaments, and then our "actuators" (muscles) attach to that structure with tendons.
Is there any existing research/samples/demos that mimic this structure?
Elastic materials are often avoided if you need either linear movement or accurate position control.
Elastic materials are favored if you need shock absorption or no seal system.
Both hydraulics and pneumatics suffer from incredibly shitty thermal efficiency. Electric engine can be 99% efficient. Any single pair of gears or a chain drive can be 99% efficient. Typical hydraulic actuator is something between 70 - 40% efficient. Pneumatics are as bad or worse.
They do use rubber air bellows in paper machinery, but it's being replaced by rigid hydraulic cylinder actuators. With pneumatics you can have somewhat accurate force (== pressure), but you never know the accurate position. With hydraulics it's the other way round. As a result, hydraulic actuators can be mounted in any position regardless of gravity. So you can use the same machinery in various positions.
why would you not know pressure with hydraulics? seems like you could detect back-pressure at the pump fairly accurately. or is that not a cheap thing for e.g. the electric motor driving it to support?
(probably-obvious disclaimer: I have zero experience here, so I truly don't know)
I was cutting corners a bit. Here's more in depth:
You can always read very accurate position with pneumatic actuators. You just can't keep it very accurately in place in dynamically loaded situation. You can also read the force of hydraulic system, but you cannot keep it constant as easily in dynamically loaded situation.
If you want (roughly) certain force and you don't care about precise position, then you have lot's of advantages from pneumatic system. The whole volume of the system acts as "hydro-pneumatic accumulator". Which means that the whole volume of the system is trying to average out the change in force. And you get that free of charge. And also you don't have to invest any money into hydraulic fluid. Air is free and available (almost) everywhere.
If you then insist on hydraulic system, you have to put separate hydro-pneumatic accumulator. But because almost any hydraulic fluid is heavier than any given pneumatic fluid, you have more mass in any given pipe section. That means the system will react slower to any(?) pressure increase. You can read the pressure very accurately, but you cannot smooth it out as fast as pure pneumatic system would.
There are caveats though, sound can travel very fast in hydraulics. And turbulence can eat out pressure very quickly. I'm kinda professional in the business, but still starting out.. :D So take this with pinch of salt.
PS. Hydraulic systems can handle 60 bar safely, while 6 bar pneumatic system can already explode and be quite dangerous. That's sometimes enough of a reason to go hydraulic.
Aah, gotcha. yeah, maintaining (near) constant pressure would require a lot more work / faster reactions because hydraulics are incompressible. That makes sense. Thanks for the summary!
One major reason I've seen is that soft actuators are complicated to use. You can't move things to a predictable location because they'll bounce, stretch as they age, and any weight changes (which applies all the time to all double+-jointed things, as e.g. straightening your elbow increases the force needed at your shoulder) change all of the above. The continuous feedback needed to counteract this is expensive and imperfect (as are we).
Existing servos / stepper motors / etc obviously do this too, but usually to a much smaller and more predictable degree.
Obviously it still has uses, e.g. as a grasper at the end of a normal robot for a softer grip. But making a whole robot out of it is still largely insane, unless you have esoteric needs. Hence ongoing research, but not application :)
I always thought this kind of stuff would be the killer app for machine learning. Products slowly drift out of tolerance, and the ml applies corrections to get back into tolerance. Apparently there’s just not enough data to respond well.
You don’t even need to get that “modern”; feed back information and use some classic control theory and you should be able to have a system that compensates for change in tolerances.
It's difficult to couple hard and soft components. The difference in stiffness creates stress concentrations that lead to the interface failing. This is also why rubber pads on laptops always tend to fall off. Creating a gradual change in stiffness is possible, but is difficult to accomplish. There is also interest in soft robotics research to make robots entirely soft. The reasons for doing this is that a soft robot can potentially not injure a human being if it's entirely soft and to make robots that can squeeze through holes. I'd also like to argue that much of the interest in soft robotics today is stems from DARPA's Chemical Robots from 2008[0], a big goal of which was to make robots capable of squeezing through holes, which has obvious military applications.
It's notable that hard and soft things in nature are bonded together through progressively stiffer bonds. Eg. Muscle to tendon to cartilage to bone. Each of those transitions is more of a 'fade' than a sudden edge.
It depends on the 3d printer. Object Connex 3d printers, which use an inkjet to spray resins of different stiffness can do this. A number of different stiffnesses are possible by mixing different hard and soft resins together. Really the only limitations to how sudden the changes in stiffness are with this approach is that the manufacturer restricts you to only nine different material types. Oh and the manufacturer won't work with you ever again if you hack the printer drivers. This approach has been used to make soft robots with graded stiffness too[0].
Pneumatic artificial muscles were inspired by human anatomy and apparently invented way back in the 1950s [1]. When I read about them in the late 1990s, they sounded interesting, but I don't think I've ever seen them in person. It seems like adding an air compressor to a robot is a big pain and most roboticists prefer to make do with electrical actuators instead. Groxx raises an excellent point in a sibling comment about unpredictable movement with soft actuators.
I just learned from that Wikipedia page that a version of the Shadow Dexterous Hand is actuated with pneumatic muscles [2].
A soft-robot gripper is perhaps more akin to an octopus tentacle than a human finger. Maybe the difference is that a skeleton is mainly needed to lift heavy weights against gravity, and all the soft-robot applications we've seen so far have involved light loads that an octopus could handle.
Hmm, I wonder how well soft robotics could adapt to heavier loads. Even with skeletal components, grip/lift strength is limited by how much pneumatic pressure the silicon can take before bursting.
Sunfolding (YC S17) is mixing rigid parts with soft pneumatic actuators to build solar trackers. This a real commercial product that is being deployed to customer sites.
The gel printing method looks very promising for printing other 3d objects. Current systems require support material or special designs without overhangs. The gel approach seems better in nearly all respects.
I'm quite impressed by this. The integrity of the parts seems quite good.
I've filed a patent on some 3d printing methods similar to this, but with a focus on ultra quick printing. Once people understand how fast these things could actually work, many things will change.
I'm thinking of the analog to the human body (or animals) ... we have a skeleton, organized and connected via ligaments, and then our "actuators" (muscles) attach to that structure with tendons.
Is there any existing research/samples/demos that mimic this structure?