Robot paper is live!
Finally! After years and years of trying to get this published I finally have the paper out in the world. Okay, to be fair it was accepted about a month ago and I did a deep dive into the paper itself (here), but it feels a little more official now that it’s out there in the wild, or at least open-access like it should be! Granted I’ve told this story before, but indulge me as I recount for the final time why this paper almost killed me, literally.
For anyone who wants to skip the sad story, feel free to jump to the end with the link to the paper.
It feels like forever ago when I started doing research, in the publishable sense I guess. I mean it feels like forever, but it’s really only been seven years or so. Which I guess is a lot of time to be doing anything, but from where I started to now feels like a much longer journey. Being early on in my career I had assumed you write something, you publish it after some back and forth, and that is the end of the story… wow was I off base.
In a lot of ways that isn’t too far off from how it happens. You do an experiment, you write about what you found, and you go back and forth with the publishers to get the best written version of the paper possible. My first few conference papers were very straightforward and I think in a lot of ways made me feel like I was an actual researcher, like someone who’s work mattered, if that makes sense. So when my then PI offered me a chance at publishing a journal paper, of course I jumped at the chance.
I really felt like the work I did was good, I felt like it was important, and I felt like it would be obvious to others how important the work that was done was. Maybe I had an overdeveloped sense of self worth or thought I was more important than I was. Maybe the work I did wasn’t as impressive or important as I thought it was. No matter the case, it was my work and I was proud of it. I thought it was worth publishing and my co-authors convinced me that this was the case. The work was what we call “high impact.”
The first rejection hurt.
It wasn’t even looked at, it was desk rejected after about six months of sitting on a virtual desk. Which is a fancy way of saying the editor of the journal read the synopsis of our paper and thought it wasn’t a good fit for the journal. That I could deal with because what can you do? They didn’t think it was a good fit, so that’s not a me problem, we just didn’t select the correct journal. That was easily remedied and we took time to revise the paper a bit, but we did and after a few months we submitted to a different journal.
The second rejection was worse.
At least this time it made it to reviewers. The reviews were scathing through and I started to question my own competence. I deal with imposter syndrome like so many of us in research/academia/life, but this really started to reinforce the fact that I was a fake and should really just not be in research. The reviews were that bad. Like, we didn’t accomplish anything worth calling engineering, much less something worth publishing bad. Still we pressed on and tried to further clarify what we did and why the paper was worth publishing.
The third rejection was too much.
The paper started weighing on me and it wasn’t just the paper. At that point it had been a year and a half into my PhD and not only was I being not so subtly told that I didn’t have what it takes by reviewers, I was also in a whole new field doing things that I had no idea how to do. There was a lot to learn and it really felt like there was no hope for me. It started taking a toll on my mental health. But we pressed on.
The fourth and fifth rejection were just a blur.
By this point I felt like I was just going through the motions. I now had two other papers that I needed to write on top of this paper. Stuff was coming in, nothing was going out, and the common denominator was me. This was my fault. I wasn’t good enough and it was being made clear. Each time we revised robot paper, and each time it wasn’t enough. Everyone wanted more, some didn’t think we did anything, others thought the knee would’ve been good enough alone, so we were getting seriously mixed signals. At the time I couldn’t even hear the good responses from reviewers, the bad reviews had buried me alive. I was years into my PhD and nothing was going right.
I stopped counting the re-submissions at this point.
Three years into my PhD and four years since I wrote robot paper and nothing was getting better. Each rejection my co-authors reassured me it wasn’t my fault, but it felt personal. The critiques felt less like critiques and more like attacks. Some of it wasn’t as bad as I had taken it at the time, but some of it was legitimately mean. At that point I was trying not to drop out or just kill myself. It didn’t feel so dire at the time, but looking back it was a very dark point in my life. I know academia is hard and it’s supposed to be challenging. I can handle challenging, I like a challenge. This just felt cruel and I don’t think anyone saw it, I didn’t even see it until after the fact.
Thankfully the dam broke and paper after paper got accepted and published. I went from having four papers on my plate to two basically in just a few months time. Then we submitted the third (last paper, which is currently undergoing the first round of edits). Robot paper was always in the background. We edited it, submitted it, redid it, resubmitted it, and repeated this cycle until very recently.
It was finally accepted.
Several of the figures I redid prior to publication because I’ve gotten a lot better at making figures, but the paper, for the most part, was frozen in time. We didn’t do more, but we clarified a lot of the things we had done. The dozens of edits to the paper did improve it I think. It’s probably clearer now than it was when we first submitted it about four years ago, but it was a journey to get from there to here. It’s only after the fact that I can look back and see just how detrimental the experience was for me. I’ve “only” written three other papers and co-authored two since robot paper, but none have had anything happen that was close to the experience I had with robot paper.
Maybe it was bad luck. Maybe it was a learning experience. Maybe it really wasn’t that big of an advancement. Maybe the publication process wasn’t as bad as I think it was. There are a lot of maybe’s, but the truth is now that it’s over I don’t really care why it happened, just that the work was finally accepted as worth publishing.
After all of this, I wouldn’t be surprised if no one cared about the result or used any of the things we designed for prosthetics/robotics. It’s been four or five years now and I’ve moved on from it. I’m doing other things that I hope will be better received. So far so good. I’ll always have a love hate relationship with robot paper. I love what I did and it was probably one of the coolest things I’ve ever made. I just hate that it wasn’t enough.
In any case, we’ve made the paper open access, because all science should be. So if you want to read about the thing that tormented me for almost the entirety of my academic research career, well then you can, but I wouldn’t blame you if you didn’t.
Experimental Verification of Kinematics and Kinetics in a Biomimetic Bipedal Robot
Congratulations! If I lose a foot/leg, I’ll know what to start designing from. Kinda serious. 😀
Quick browse through sure looks like it should be a touchstone for new designs and a useful joint system.
LikeLiked by 2 people
June 6, 2022 at 8:53 pm
Thank you! I would probably design my own prosthetics if I needed one too haha!
June 8, 2022 at 3:51 pm
Now, now. It was enough. It just took a while to find a set of reviewers who could actually see that. It got published. Don’t turn self-deprecating now that it’s finally out. (yay!)
I’m admittedly only a hobbyist, but from your thesis and blog overview I found the knee genuinely interesting. Makes me want to think about how I could try something more biomimetic for quadruped joints. (Someday. I need to figure out the dirt simple version first.)
Now I have to recall all the questions I had, see if the answers are in the paper, and ask you if not. But I’ll probably have to do that later, because it’s so late already. (Where did this evening go?)
LikeLiked by 1 person
June 7, 2022 at 1:07 am
That’s fair, I’m happy enough to have it finalized and out there, I don’t care if the other reviewers weren’t impressed. It was a bit traumatic though, the experience does make me question if the stuff I’m doing now, in particular “big idea” is actually something impressive or if I’m just fooling myself. I guess time will tell.
I think biomimicry isn’t too hard to do. It tends to give a better result (in my opinion anyway) and you can do it on the cheap. I think the big hang up is time really. Good luck finding the time 🙂
No rush, I’ll be around to answer questions whenever you need!
LikeLiked by 1 person
June 8, 2022 at 3:34 pm
Having now read the paper, I am back with all my questions. You’ve talked about the knee a lot, so most of these are about the other joints.
You used a pretty sophisticated 3D printing process to get high-strength pieces (I think I remember you mentioning that you got access to the printer just for this project, and it was expensive). Did you have (or expect from models) problems with things breaking, or was this more of a proof-of-concept for making parts with the longest possible wear life?
I was curious about how you were planning to constrain the motion of the synovial joints, since some of them have more degrees of freedom than the corresponding human joints. The paper answered this when describing the ankle joint: the air muscle sleeve is difficult to twist and therefore doesn’t permit much rotation, so that prevents the ball-and-socket from realizing one of its degrees of freedom. I’m still wondering whether this might yield a looser motion than what you would get out of a biaxial joint, or whether taking away the ball-and-socket’s rotation makes them basically equivalent.
You mention that joint stiffness in humans is dynamic (makes things more complicated for robot designers, but I think that’s neat). And you could replicate this in the air muscle joints using both their passive properties (stiffness changes as the joint is bent) and active control over the amount of inflation. And that makes me wonder, is the human version just passive? Or are there active elements too – like muscle pairs that aren’t actually driving the motion tightening to provide more support, maybe?
Inflating and deflating the air muscles around the synovial joints only adjusts the stiffness, right? It doesn’t provide actuation of the joint. I’m curious what your team’s future plans were for adding actuation, if you ever got that far. More artificial muscles? Were you planning to add tendons so you could mount the actuators off the joints?
Your minimum test pressure was 20 psi, looks like. Is there any particular quirk or disadvantage that comes in if the air muscles aren’t inflated at all? I’m thinking about whether one could get some of the benefits of the artificial joint capsule without including an extra pneumatic system.
LikeLiked by 1 person
June 13, 2022 at 12:49 am
Also, since there’s a sliding component to the joint, how does the increased coefficient of friction affect use and longevity. One of the huge advantages of ball bearings is their effective CoF is like a human joint’s without the upkeep organic joints require. A well greased or actively oiled polished metal surface would be pretty good, but even a solvent smoothed 3d printed part is likely too rough to have a good service life.
LikeLiked by 2 people
June 13, 2022 at 1:07 am
Oh great question! The joints were designed to wear to the kevlar, which has much better wear properties than the nylon/carbon fiber composite the outer layers are made from. We could theoretically lube them via the pneumatic system, but that’s a whole other set of tests. The lab (as far as I know) is still trying to figure out the life cycle of the joints, so we’ll have to see how they perform. Mostly this is for experiments though and the prosthetic side of things would be made from metal or similar materials, so the CoF would be much better! Speaking of which, the knee is currently being tested for prosthetic and was made out of metal, so kind of cool seeing it progress like that!
LikeLiked by 1 person
June 13, 2022 at 7:14 pm
All great questions!
The printer was roughly $15,000 at the time (if I remember correctly it’s been ~6 years or so now) it’s available to the public, but at that price point… oof (actually they make cheaper versions with less capability now, I just took a look at the website, still pretty pricey though). We bought it because prototyping the parts was a lot faster than when I was stuck machining them and the geometry could be far more complex. The reason we went with the Markforged (name of the printer) in particular was because we wanted strong, ready to use parts. Ideally we wanted the robot to support 100+ pounds so a regular 3D printer wouldn’t cut it. Since we could add things like kevlar to the print it also gave it very good wear properties. It’s literally imbedded into the material so you have to wear through the plastic (nylon/carbon fiber chop really) to get to it and we haven’t had that happen yet, but it’s nice to know it’s there if we do ever wear through. Basically we just wanted something that was ready for testing and not just a prototype.
So rotation is an issue, which was sort of the major limitation with the design, in particular for the hip. We can fix that with a custom-made sleeve, theoretically anyway. In the end I didn’t want to constrain the movement. I mean the PI will add blocks to the parts to constrain them when they are being tested like that, which I’ve designed for before I left, but I really wanted to mimic humans. The thought is that by not constraining the motion it would naturally adapt to rough terrain. Having a little “fuzziness” to the location of the joint in space makes it a little harder to control traditionally speaking, but my working theory is that it would be much more human and self-correct in a lot of cases. I mean it works for us, so I was working under the assumption it would work for a robot too, if that makes sense.
Yes, inflating would only change the stiffness of the joint. Actuation would be done by attaching more of the air muscles to the robot like our muscles are attached (tendons like you suggested, great guess!). The lab actually published another paper a year or so after I left looking at modeling the correct placement for this purpose, which I dug up if you’re interested! https://link.springer.com/chapter/10.1007/978-3-030-24741-6_2.
Ironically I realized after the tests were done that we never checked with ambient pressure, which probably would’ve been much closer to human than we got with the 20 psi. There’s no reason why we couldn’t do it and in fact, for the robot photo in the paper the joints are not inflated. So yeah you could avoid the pneumatic system all together if you wanted. In recent years people have caught on to the idea somewhat of using a spring like joint, but none of the designs were as simple as my synovial joint solution (mostly fancy rubber band designs).
Some behind the scenes facts, since you’re here:
(1) I was in France when that data was collected, so I probably should’ve clarified to the person I left in charge of testing the joints to test without inflation, but hindsight is 20/20.
(2) I rushed the build for the test rig for that and the photo in the paper is awful because I took it to show how to use the rig. I never planned to use it in the paper but the person who did the tests didn’t take any photos for me!
(3) If you look closely at the robot photo, there was a trashcan in the background that I had to photoshop out (my co-author really, really wanted it gone). I feel like it’s super obvious since I’m horrible at photo editing, but no one ever said anything about it so now you’ll notice! haha
(4) I also only took that robot photo for me. It was the first time the robot was fully assembled and standing. I was (and frankly still am) in awe of the fact that it was self balancing like that. It was a good day.
LikeLiked by 1 person
June 13, 2022 at 7:10 pm
They thought the trash can was too offensive to be in the photo? That’s so funny.
I have some mental scar tissue about imperfectly constrained motion due to my history messing around with the quadruped. I would think “It’s okay if this joint has a little bit of ‘slop,’ I’m not aiming for clockwork precision here,” and then I’d get to the end of the process and the robot wouldn’t even stand up straight, because a little out-of-plane deflection in each joint turned into major instability and tilt across the whole linkage. You already did better than that, because your bot could stand unpowered – so the damping from the joint sleeve must control that sort of thing. Hopefully in a walk cycle it does just as well. I do very much like the idea of having compliant elements in the design.
LikeLiked by 1 person
June 13, 2022 at 10:57 pm
Well he basically thought it looked bad. I really wish I had taken the “professional” shots I meant to take, but life gets busy, you travel to multiple countries, then have to drive cross country to start school the next day you arrive to your newly purchased home you bought sight unseen. I mean we’ve all been there.
Oh I can see why you would be a little anxious about my robot then! I think the compliant elements help constrain things enough that it’s stable “enough.” I read somewhere that from a control systems perspective humans are marginally stable as it is, so I assumed if I could mimic that slightly then good things would happen. Of course, it could be a nightmare for the lab, I don’t know, I haven’t heard much about it since I last spoke to my former PI about it and progress has been slow since I left from what I was last told. My Solidworks files were a nightmare of a mess, so I do feel slightly guilty about leaving it that way for them!
LikeLiked by 1 person
June 14, 2022 at 6:47 pm
Sorry, I can’t seem to reply to the proper response. I like the idea of wearing to a substrate. I’ve seen it before somewhere, but I can’t seem to recall where.
Reminds me of a random story about the T-34. When the Germans built a tank, it ran from the factory. The Russians had such a rough machining process that in the first few hours of operation it shaved a kilogram of steel out of the engine and dumped it into the oil pan.
I love practical solutions. Self-leveling and self-wearing structures are as practical as you can get. Of course eventually you get to wear enough you’re back the pin joints being load bearing, but as you say, in biomimetic designs eventually you have to use ligaments and tendons anyway, so . . .
LikeLiked by 1 person
June 14, 2022 at 4:46 am