How exoskeletons can respond to your motor commands (Part 3 of 3 on exos)
This is Part 3 of a multipart series on exoskeleton research. If you haven’t read Part 1 and Part 2, I’d recommend going back and reading those first.
At this point in our story, we want to figure out how to stop making an exoskeleton feel like a “separate intelligence” strapped to your body (like a 3 legged race partner), and more like an extension of your body. Let’s imagine that we successfully implement the motor command sensing system from Part 2, so the exoskeleton knows at all times exactly what you want to do with your body. Now, what can the exoskeleton do with that information that would be helpful, but still feel like an extension of your body?
A professor of mine once told me something to the effect of “the trick with research is to avoid building bridges into the ocean.” What he meant was, if you’re going to work hard solving research problems, make sure you know how it’s going to connect up to something useful on the other side. My goal with this post is to increase your confidence that if we build the motor command sensing system from Part 2, we have ideas for what to do with it — that we’re not building a bridge into the ocean. This entire post is (by necessity) speculation, because we still have to finish building our bridge from Part 2 before we can test anything. Because it’s speculative, I’ll try to keep the post short, and just briefly gesture at a couple of ideas. The point is, there’s land over there!
The simplest idea is to take the motor commands we sense from the pilot, and apply them directly to the available motors. For example, if the exoskeleton senses you bending your elbow with 2.3 Nm of torque and we happen to have a motor attached to your elbow, we could (nearly instantly) apply 2x that value (4.6 Nm) of torque at your elbow, and triple your arm strength. If at the next moment your elbow relaxes to -0.1 Nm of torque, we could update the motor to apply -0.2 Nm. Running that in a perpetual tight loop may feel great, or it may feel terrible and unnatural. I’m excited to find out!
If that feels unnatural, it’s probably because it is unnatural. Humans evolved to rapidly adapt to a whole range of changes in our physical environment (e.g. putting on a suit of armor, carrying a heavy rock, wearing a new pair of shoes, moving in waist-high water, etc). Humans didn’t evolve to adapt quickly to having our joint torques magically amplified, because that never occurs in nature, and so it’s possible that that just doesn’t feel right when an exoskeleton does it.
If that happens to be the case, we can explore doing “physics based assistance” instead, which might be more intuitive. Imagine we wanted to use the exoskeleton to simulate the weight of wearing a heavy backpack — I know, we wouldn’t actually want to do this, but bear with me. We could simulate the sensation of wearing a heavy backpack by applying carefully chosen torques at your hips, knees, and ankles, bending your joints as though you had a heavy backpack pushing down on your back. If we did a good job simulating the backpack, you would instantly adapt to the sensation, because your brain evolved to handle things like wearing backpacks, and you’d be able to walk comfortably. Now imagine we wanted to instead simulate a “negative backpack,” by simulating that your torso had less weight on it than before. That is also just a matter of carefully computed torques on your joints. Maybe your brain would adapt to that just as quickly as the “positive backpack”? Hard to know without running the experiment, though some prior art exists for simple upper body exoskeletons. There are all sorts of “simulated body modifications’’ we could do with the exoskeleton, like changing the perceived mass of your limbs, the strength of your muscles, the strength of gravity, etc. There are an infinite number of potential controllers to explore. I don’t know if any of them will feel right, but it’s worth a try!
If neither of those controllers feel really good, I’ve got a few other ideas we could try, but I’ll leave it at that for now. This is all raw speculation until we get the motor command sensing described in Part 2 implemented and out in the world. I hope I have at least left you with some confidence that if we do get motor command sensing to work, there will be fun exoskeleton control strategies to experiment with :) Hopefully I’ve even sparked a few ideas of your own. If so, please get in touch!
