Alexander Badri-Sprowitz & Monica A Daley "BirdBot"

[MUSIC] One of the things that I think surprised me is when we were doing the actual dissection of e-mute cadavers and we went back. We had the model of the system when we've sought. We, you know, well there should be this tendon linkage that goes all the way up through the leg to the femur. And so can we look at the knee anatomy and see evidence of these connections? It's incredibly complex but you could actually follow the connections between the connective tissues all the way up to the femur despite this complexity. So when you have that hypothesis, you can go and look for the essential elements that are there. It's true that they have this complex knee structure and yet you can identify these linkages of connective tissue that connect all the way up to the femur as expected. Any model has simplifications, right? And this is the constant challenge of biology is finding the right level of simplification for the question you're asking at that point in time. We do not want to rely on something which is not a point foot because something which is not a point foot is hard to model if you have an actual foot, a second-ended foot and you have a changing center of pressure. That's nothing so simple. And if I have springs in my system and maybe dampers, physical dampers in my system, then again I need to model those ones, they can react on the own, they have their own dynamics. And so they complicate my model-based approaches. And that's absolutely true. How biosepited robots are being controlled, like literally what kind of controllers are implemented, they are quite different. So I agree totally with Monika, it's a matter of what is your task? Do you try to solve a technical problem as an engineer and you want to have a robot-elected robot going up the stairs? Or do we want to try to find a model and explanation, a proof of concept of what we observe in nature and can we actually have this proof of concept in a hardware robot which creates an often-advanced slide robot. In this podcast I'm sharing my passion and curiosity for software botics, where we share inspiring stories about the work we do and how we can wash the limit. I am Marad Twainy and this is Software Botics Podcast. Support for such a show come from Science Robotics Journal. I really find science robotics to be a great resource for reliable and tangible research, where we can really push the limit of the science we do in robotics. Great way to stay up-to-date with the published article is chicken out the released monthly issue. All the links will be included in each episode description. We will also have a regular conversation on the most published science robotic articles, where also you can contribute with your questions and thoughts about their research. Thanks Science Robotics for sponsoring Software Botics Podcast. First of all, a conversation for the paper. I think it's very interesting, the bird bot project. I think it's maybe I want to ask at the beginning for Monica because I think the concept of an inspiration from ground, bird and how they use morphological intelligence. There's no sensing and actuation. I think that's very impressive when you speak about how we can design robots or soft robots for an inspiration like that. Maybe that question for Monica at the beginning. Can you tell us more about what's really fascinating about studying birds because there's many different classification if you can tell us and what's actually interesting about the ground robots that can't flight. I've been studying ground birds for I think about 20 years now and what's particularly fascinating about birds to me is that they have a completely different evolutionary history and morphology from humans. We're most familiar with you know biomechanical models of human walking and running, right? But humans are a very unusual biped with the only you know species that walks with this completely upright posture. There's many different many morphological differences in birds and they have a long evolutionary history of bipedality. So all living birds are bipeds and they come in a wide diversity of body size as well as you know locomotory ecology and behavior. But ground birds exhibit a form of locomotion that is thought to be quite similar to their ancestors among the Theropod dinosaurs. Some of that the the overall architecture of the leg, the number of segments and the muscle tendon systems that we see in running ground birds has a long evolutionary history that goes back to dinosaurs via Theropod dinosaurs. And this the diversity that we see in in birds I think reflects that some of the features are highly adaptive in that they allow robust and stable locomotion over a wide variety of terrains with relatively simple control. And my research in biomechanics has led to many hints about the specific how the specific morphological mechanisms allow for this intrinsic stability of and in mechanical sort of intrinsic mechanical control. But it's very hard to directly test hypotheses about this in living animals because many times you can't make all of the direct measurements you want to make and you can't remove the nervous system to demonstrate that it's fully you know mechanical control. And that's where the collaboration with engineers in building these robots is particularly powerful. It's that we the physical the robot is a physical model of our understanding of how the bird system operates. And it's of course more simplified than we see in living birds. But at very nicely demerit straights the overall function of the the lower limb and the function of these tendon networks in providing intrinsic stability and elastic energy cycling for economical control with very simple mechanisms to allow stand swing transitions without a lot of sensing feedback. What I really like was project using robots to understand what's actually happening at the mechanism. That's really impressive. Maybe I want to ask Alexandra in that case. I think you have this great experience in designing the robot. But I want to ask you across the the project you have been doing many years now. How do you think about physical intelligence because clearly here you have going for concept how we can't demonstrate more vertical intelligence in the design. Can you say something with it missing in this the design of ligged robot locomotion. And yeah if you can tell us more what's interesting about the design to get for physical intelligence and just let's reduce sensing and actuation. Yeah this is a super interesting question as in it really drives much of the research of the entire field. I think physical intelligence allows to extend what is currently in what we control in robots directly. So often we in a ligged robot we want to have exactly the amount of torque and displacement in the joint which we command. And as an engineer you will build the system accordingly. But what you don't control is the environment and what you don't control is noise in the system. What you don't control are deflections which are sometimes inherent coming from aging of the of the parts. And all of these unknowns can be captured if you have a design which can can can react or it can compensate in one or the other way for that. So I think physical intelligence or morphological computation and there are many names I think behind that. It allows you to embed some of the functions into mechanics. And then from that and we see this in biology in fact you this these components can react often much faster or they can be smaller and they can have higher power outputs or they are inert to communication delay. So there are many advantages and there are many different concepts and mechanisms which have been found over the many last years. And we can draw quite many out of the box nowadays and we can input them into robots. Maybe I want to ask you again because I think in the in the field for robotics maybe soft robotics the idea that let's go for reduces sensing and computation and actuation. And in this project I think it's very very interesting to exhibit that. Do you think this is really sufficient? I'm not like to say that these are just maybe simplified representation what's actually happening in the locomotion for birds that cannot fly. Do you think this sufficient because you already demonstrated that there's obstacle they can really there's no delay to adapt. But do you think journal is speaking that could really just say let's reduced completely sensing and actuation and robotics or that's something specific here. Yeah, so I would not, I definitely would not say that it's sufficient like to have the the birdbot style clutch like design, right? So that is a what I consider and I think also Monica as a base function of this like so it allows to to produce a large range of locomotion patterns and robust engagement and disengagement and so we produce like forces and all of that. And so it's basically covering, I don't know if you have 100% that maybe it covers like the 70 80% of the basic like function, but it does that in a with a mechanism. It's a not not a controlled system. It's a mechanism which engages through. There's still a field forward control part behind that right and so it's it's it's it's clearly not something which you want to rely on completely if you want to have a robust. Freedom, actually running animal or robot and you need to accelerate need need to decelerate or if you need to hop around and if you need to be adaptive to all of this points or if you need to react right so if you. As a bird if you see a predator you need to run away you can't cannot only rely maybe on one set of muscles and that's what we see right so you don't just look at one extensor set of muscle you have many. Activators inserting into many joins and the the function what we describe is the major function for which is one function in leg and locomotion you to carry your body. So you want to carry a body and so while you can put a parallel spring into into into into the robot and also something like this it's possible at least in the bird. But and you need to take that spring out of the system for swing phase but it's not the only function of like a locomotion right so it's not only about carrying you also need to hop in so all of the examples I give before and for those ones you definitely absolutely definitely want to have. Control you want to learning you want to again you your system changes right so you're aging the animals are aging or you loaded you carry a backpack or you're sitting on a horse I don't know whatever it is or you're like you're eating you're belly gets full and so all of that has to be absolutely captured by. Reflect systems by sensor feedback by by like close to control like animals robots whatever you want to talk about. But I just add to that before we move on I think one of the the nice things about having this mechanical control element is that it's specifically taking care of those elements that have to occur very rapidly so there's unexpected changes in the ground contact conditions and the intrinsic mechanisms in this robot allow those. Transitions between stance and swing and swing and stance those transitions are very important to maintain stable dynamics the intrinsic mechanics handles that part which happens on a very rapid time scale and this allows the control to be applied on longer time scales so to regulate foot positioning from step to step for balance and while we haven't implemented more complex control in the robot we do. You know we do think that you could then layer on these to get even more agile and robust control mechanisms right by combining the intrinsic mechanical control with feedback control and we do see lots of evidence that this is in fact what what animals do right they do they have substantial transmission delays in the nervous system so in when they're running rapidly the fastest response time. Is a large fraction of the stride cycle so they can't actually implement feedback control immediately they have to rely on the intrinsic mechanical response. You know allowing for to to to keep the performance within a certain range before the feedback can then come into play. Maybe I want to ask you since you have been working for now more than 20 years when you look to the ground birds if there is something maybe very interesting when you compare with other species and especially if you speak about the flying one and not flying do you sing having flying capabilities with a fixed locomotion because I think it's very interesting that they can stand even this leaving and there is no energy consumed and that's fascinating how this happening. Can you break down detail from an inspiration when you try to study understand clearly this project seeing the the robot try to explain that when you try to model do what is the most significant part on the animal or the bird here the ground bird to understand what is actually happening here. I think from my perspective understanding two elements is really essential one is the geometry of the scuttle lever systems and understanding how the the muscle tendon systems are acting at each of the joints. The careful gearing of the different joints in the limb is how you fine tune the motion that you get upon loading of that leg and then the other element that's really critical is the the tendon network that is a multiarticular tendon network that is carefully designed to then transfer energy among the joints in a way that allows for the limb to be stiff and store and return elastic energy during stance but to be rapidly shifted through this by stable tendon acting at a by stable joint in the lower leg to rapidly shift into a flexed position which then makes the whole limb slack and very easy to flex for the swing phase. So it's the combination of the geometry of the skeletal system and the action of these multiarticular tendon networks on those joints in the system. So I will most repeat what Monica saying on this one right but yeah really the individual task which we needed to tackle was in the animal you have segments of bone segments and you got a patella and other sesame bones which act as the lever on for your muscle tendons onto the joint so that you individually produce torque in each of the joints but we know in words you have multiple leg joints and you want to coordinate really how much torque is being. Induced in each of those joints and so the basic idea was initially to say to have one common tendon so basically one mechanical coupling which really induces by it because it's a tendon it's a it's a serious element has forced it's equal force everywhere. That means we want to induce but we want to use different amounts of torque for each of those joints because each of the joints has a different distance so here but I'm painting a little bit is the geometry which is which we really reduced in the robot so we made much more simple but we had to understand that in the animal you can have still the representing components the parts as in bones muscle tendons. The the lever arms of the of the tendons and we needed to have this transfer into a simple map presentation to eventually so because that was the other part if you if you don't assume that simply simplification and we try try that before I actually found it was very hard to solve if you without the simplification so once we simplify it into the robotics setup and saying okay we using linearist can mechanisms using. A pentograph lecture is the base structure underneath and we can create a single tendon network which produces a programmed hardware program amount of joint toward for each of those joints so this was this this first this muscle tendon network to have this transfer so that that was very important for us to understand and the second part is we want to have this model this individual this single muscle or spring tendon network now and we want to switch it on and off right so we want to have this by stable. We have it is the from the outside and really they're looking at the bird and seeing that those most distal joints the toe joints and the bird that they would. It flex and extend so much that we could use that action at that motion to if we replicate it in the robot and we could therefore see how much tendon gets lose or not that we could use it actually to switch often on the more proximal network that was the second part which really was needed to to put it on. And put those two things together and there were other smaller items like tendons like and so on which we also wanted to understand and it was really necessary to it was really helpful to actually go back also to the bird and with Monica and really look again and say okay we think there should be something and so it's a big puzzle was a big puzzle in fact. Maybe I want to ask you here again both of you there's something when you try to come up with understanding what's actually happening and there's something maybe Monica or you this really serve priding or maybe not really straight forward is priding or not really explicable. Do you have this kind of discussion was maybe confusing to explain what is happening at this locomotion or what. is really, since you said simplify, do you think there's extra stuff needed to embed it and the robot to capture what you see or exit-bling it? In case you want to think, I have one point. Okay. Yeah. So, yeah. So, what I just explained, right, so we simplified from bone and patella and sesame mode structures into this, into a tent network which is comparably easy to calculate. In fact, you can draw almost the mechanical conclusion of that on a piece of paper and you can geometrically solve it really. It's that, it can be that simple if you look close enough. But what still remains is the complexity in the animal, right? So, we know that the animal does not have single hinge joints, but these hinge joints they have a rotating, like the center of rotation is shifting in between two bow and surfaces, right? The articulation is not producing a hinge joint motion. And the same goes for the patella motions or the sesame motions, they're not perfectly acting as camps, but they're highly non-linear. And this has been documented many times before. And that, each of those instances you have multiple times for the entire leg. And we simplified that and we said, well, we linearized everything. But what is not straightforward is now to go back. And if you would ever want to try to see if these multiple non-linearities might add up or kind of outweigh each other to become linear again. So I think that that is something where I'm right now. I'm curious about you. Yeah, understanding the complexity of the biological anatomy is a constant challenge. And one of the things that I think surprised me is when we were doing the actual dissection of e-moot cadavers and we went back. Well, there should be this tendon linkage that goes all the way up through the leg to the femur. It's incredibly complex, but you could actually follow the connections between the connective tissues all the way up to the femur despite this complexity. It's true that they have this complex knee structure. And yet you can identify these linkages of connective tissue that connect all the way up to the femur as expected. And many of the simplifications that we see even when people try to build fully anatomically correct musculoskeletal models of the e-moot ostrich. They're still making simplifications in how they model the muscle systems and the origins and insertions. So we were actually surprised by the knee anatomy because we've had these pictures that are based on the anatomical studies and they too had oversimplified it. So based on those anatomical studies, we didn't think that the connective tissue connections that were needed were necessarily there, but when we looked in the real animal, they were there. So any model has simplifications, right? And I think we do have a reasonable level of abstraction for understanding this whole limb function and the connections between the joints and the transfer of energy between the joints, but there's still a lot to be understood about the complexity of joint function in humans as well as animals across the board. Maybe I want to scare about intelligence and design from two perspectives from your side and Alexander, but maybe you first, when you look how evolution can come up with such design like that in the ground birth, for example, and also transformation from dinosaur to flying bird. How this happening? Because now we speak about design, eventually robotics, we try to design what's morphology, what does it really decide, et cetera. When you try to look how evolution can come with this transformation from dinosaurs to bird, how this happening? That's a big challenge to address that one. A lot of many evolutionary biologists would cringe at using the word design in the context of understanding animal biomechanics and function because it is an evolved system. And for some people, design implies some intelligent, you know, creator, some intelligence designer. But evolution, we know, is a lot more complex than messy than that. But evolution is an optimizing process, right? It is not generating optimal solutions. Many of the things, many of the features that we see in animals can be the results of millions and millions of years of evolutionary baggage. Structures that have been were adaptive for function, optimized for function in a different environment and different context. And they are still there because there's a lot of developmental genetic inertia for changing these systems through natural selection. And if it doesn't actually hurt, then it will still remain. So I think of evolution as an optimizing process, but it's not creating optimal solutions. It's creating good enough solutions, depending on the specific, you know, selective pressures that are placed on the animals. So we always have to be careful about when we're trying to make arguments about a specific feature being adaptive. It may be that the complexity of the knee joint that we observe, for example in birds is adaptive, or it may be that it's just kind of an evolutionary legacy joint that has gone through lots of different shifts in selective demands over many, many hundreds of millions of years and some of those features could just be good enough for what it does. Yeah, I just want to emphasize what Monica is saying is also based on the many papers in that field that are showing that the changes from dinosaurs to birds are really gradual. And you find so many samples. We're working with like so, like posture, the crowdiness, the segmentation ratios, there are all of the intermediate steps that are available in the fossil record. And the papers are there. And so it changes, slide changes, like very small changes on the center of miles and then how the trunk shorten and change posture, how the tail got shorter. So that absolutely supports in every way. This is an optimizing process, it takes small steps, there's never a sudden, ginormous change where suddenly the head is at the end of the tail. So it just doesn't, like we see exactly how evolution is being hypothesized and it's supported with every piece of evidence, which is not my research, but it's out there and you can read about it. And I love reading and seeing those examples. I just wanted to add in terms of the discussion of evolution, evolution may be a messy process, but one of the great advantages to studying birds is that because of the diversity that you see among birds, we can look for features that are convergent for specific functions among different lineages, right? So for example, not all ground-running birds are directly related to each other. Yet we do see convergence of similar features of elongated distal limbs and lightened muscle mass in the distal limb and elaboration of tendons and greater multiarticular tendon connections among the birds, specifically the birds that are most specialized for over ground locomotion. So that's how we can think about looking for clues in the evolutionary process by looking for convergence of similar solutions among diverse unrelated animals. Maybe what question for you, Monica here? When you look to, you have been sort of different species, but if you can give us comparative we, the locomotion, which one may be more efficient since we do with uncertain environment than we, for example, they have to escape predators or there is injured. If you have the holistic image, which one be more interesting for locomotion human or animals or birds here, there is many, many stuff here to compare, you are expert here and a gentlest, which one more efficient? Well, I think looking at animals that excel at a particular aspect of performance are always interesting, right? These are the fastest bipeds on earth. So they're, and they have a large body size, they can reach high speeds, they can go for long distances and they have more economical locomotion compared to humans. So I think that that's fascinating. But it's also intriguing to think about all of the small birds that are bipedal but live in three-dimensional arboreal environments. And so they're having to go up and down the trunk of a tree. They're doing bipedal locomotion in three dimensions and this probably has some really fascinating features, adaptive features in the foot morphology, for example, to enable this that we haven't really fully analyzed yet. Maybe for Alexander, that's a question I will ask you. And to have this rich experience in many robot design, this time when we're troubled, for example here, what really changes the new perspective of that design? Which time you try to see what will be the end goal, but what really changes every time the perspective of the design process to come up with really intelligent design? And how it's easier or complex if you can't look for it more. Okay. So I had been building LEGO robots for this project specifically. And I had been implementing with a team at EPFL in Alka Easper's laboratory, LEGO designs, which we call parallel elastic LEGO designs. So there's a springy LEGO and spring supports the LEGO function. And these kind of LEGOs work wonderfully, like for example, with open-loop control or with feed-forward control, LEGO center pattern generators. And they are robust. When you run them, they don't need sensory feedback and such. They have this one caveat that in swing phase, so in stance phase they work wonderfully, but when you need to swing the leg, you need to work against that spring. And this is something when you build this as a robot, you recognize this because you look at the power supply and you see, well, my hip motors, these draw a certain amount of power. And the motor which is responsible to flex the LEGO to knee joint and to shorten the leg. And that's massive amounts of power and then you think about, yeah, that's because you have to work now against the spring. And that's not what happens in the animal. And so specifically for this project, what changed the, my perspective was joining Monika's group as a postdoc researcher and in that process being able to look at birds, bird legs, and specifically at the leg function in large birds in cadaver legs also. So I saw that I could see that these legs have a mechanical coupling, which happens in certain of the joints. So when you move in a cadaver leg, if you move that leg, then the other leg joints also moving along. And that's a consequence as I understood then of the muscle tendons which are in there. And so even if this is an inactive system, there's no neuro control on this coupling exists. And so what and that happens especially in the distal leg where we, where we documented this also here for this bird project. And so if you combine those two tasks now, if you recognize you have a parallel leg spring in the robot and that's a problem in swing phase. And then the other side you see the mechanical leg coupling, you see the movement of the toe joints or the foot joints. And if you combine them correctly, then you can basically get rid of that problem that you have the parallel leg spring. And so yeah, I think that's where things got together. Maybe for Monica, I don't know if you would like us something here. I think I remember when we did this dissection on the on the e-mail cadaver, right? Well, I wasn't there for all of them, but it's really fascinating to see when you're, you know, doing this hands-on how you can put the foot into the position that, you know, relates to the stance position, right, where it's flat against the ground. So if you're pushing it, you see you feel this direct resistance due to the passive loading of all of the tendons. But if you let the toe flex into the flex position that is comparable to the swing configuration, then everything is slack. And you can see physically the slacking of the tendons in the system. So it's seeing that physical demonstration in the actual cadavers or doing the actual dissection was very enlightening for us, I think. That's a good point. Maybe a quick question for you here. The tension, emotional evolution is messy, et cetera. But when we look to buy an inspiration and a biometric, do you think generally speaking here, if we speak at designing that problem, when you look to the ground parts here as example, do you think it's really optimum? Is something you think could be in robotics could be enhanced? I know it is not the case, but I want that to be, you have to solve that. Maybe there's something could be enhanced in robotics side. So clearly what we implemented here as a proof of concept can absolutely improve robotic leg designs. And not only that, this mechanism should also be applicable for prosthesis and orthesis designs. Because what we have is we have a compliant leg in stance phase and that compliance switches off. This is called a clutch in engineering. And this kind of clutch, clutching many mechanical solutions engineering solutions for clutches had been proposed before our work. These were engineering solutions and they often either involve control or actuators or there may be more complicated, more heavy, complex. And here for the first time, we could actually on each of those points, birdbugs clutching leg design, the clutch is super lightweight. It's literally just a tendon. And it evolves barely a few, the spring itself, which we have anyway. It's absolutely robust as long as the toes are pointing into the right direction. This clutch will always engage. It will, it's fully scalable. We can make that mechanism which we show here for a 29 centimeter hip height robot. We also show for one meter, 75 robot, like not robotic. It's just like what we built there. But the equations indicate that there is no size limit really. So it's fully scalable. And so the importance for therefore for robotics is now we have for the first time a electric clutching mechanism which outer engages, which is fully scalable. It costs no energy. There's no control needed. And that was just missing before. So we do expect a lot of changes to see in the robotics community based on that. I do expect that. I think I would just reiterate what Alex already said. So I don't know if I have more to add there. I do think there is a kind of in general in robotics there has been a tendency to have legs that have point feet, for example, very simplified feet. And perhaps this design highlights how effective design of a foot can actually facilitate better mechanical control of the system. Yeah, I would like to add to that since it's not a mention of robotics and we had an episode with you here and he was mentioning that the energy was a problem consumption. And really this is impressive that the energy here is reduced. So I think yeah, it's really fascinating. Maybe I want to go again for the injury and redundancy. If there is this damage or failure in the bird, how they can adapt this mechanism. Maybe from, well, sorry, it was your Monica. For example, when I see bird, they're in balance and I'm wondering if they use morphology intelligence. How the sensing here could reflect this, you can't elaborate more about situation here. What's happening? Yeah. Well, in the real bird robot, we don't have just ligaments, right, they're muscle tendon systems. So we have the mechanical linkages, but we also do have substantial muscle mass controlling the co-activation of those muscles and increasing that co-activation with feedback could be an important mechanism for preventing injury, right? But also all biological tissues, tendon and bone are adaptive over time. They will remodel based on the stresses actually experienced. So as long as there's not a catastrophic overload injury, these tissues will remodel and, for example, if there's consistently high loading, you would see remodeling of the tendons to be stiffer, you would see remodeling of the bone to increase the cross-sectional area. So in the real biological systems, there would be adaptation over time to fine-tune these structures to meet the typical loads experienced during locomotion. And through reflex feedback to the muscles, you can then respond to sudden impacts to help absorb energy and help to minimize the risk of injury. So we see that the mechanical control systems have to be acting in concert with the muscles that are there. The muscles would play a very critical role in preventing injury and responding to the unexpected perturbations. Yeah. So I like Monica's explanation of considering, of course, again, the physical mechanism on one side and then sensory loop base or reflex system to have close loop control in some way. And of course what we also observe in the animal there are several muscle tendons applying torque at joints and then over multiple. So it's not just one, you often have multiple ones which are inserting and they're producing forces often during different parts of the load cycle, the gate cycle. But you could also have like muscles which are coming not exactly in that plane where the major muscles acting but maybe slightly off plane. So this concept of having multiple actuators kind of doing the same job, we don't apply that in robots really. Because we have one function as an engineer, I have this one function and if that's the function and I don't need to add a redundant system but because legate robots really rely on a low body model and you don't want to add more mass and necessarily to keep the legate robot as dynamic as possible. But of course if you see the birdbot design and if you now imagine a future version of this robot which has a sensory loops on top of that and it would have maybe a second actuator or in this case a third actuator and you add this redundancy, this would help of course. So if something, yeah I don't know, the point is if you cut the tendon and the robot the robot will not work. So it's like redundancy wouldn't help, it's just too simplified I think. But yeah if you replicate what we see in animals, if you would put many more actuators into the system, you could clearly achieve similar redundancy effects. So maybe I won't ask you again if there is disagreement about the way of the design legate robots, maybe from a market perspective generally we should try to see the explanation or hypothesis for locomotion and for you as well, maybe disagreement in that approach or explanation generally speaking in the field. I think there's some disagreement in the field about whether or not it's actually useful to use bioinspired approaches or whether we should instead be designing based on fundamental principles and engineering design. And I think our robot is a nice demonstration of how you can get innovative features in leg design through understanding the biological system. But I see birdbot more as a physical model of the biological system that allowed us to test hypotheses about biological function. From my perspective as a biologist, that was the most fascinating part to me. So whether or not it is the right way to design robots, I don't really personally care because it is the right way to test a biological hypothesis. So maybe Alice can chime in on the controversies in the robotics field a little more. Yeah, I agree with Monica the idea of using bioinspired designs by mimicry is being discussed. I think there's enough evidence, enough papers out now that clearly show advantage of certain mechanisms. And these ones as a robotic design and you have to make the choice if you want to incorporate them or not. We do see a strong tendency right now to have legate robots which use few mechanical elasticity and they are instead directly actuated each of the joints directly actuated by a strong actuator. Brushes DC motor control and they're very strong, they're very powerful. And the reason for that is that over the last 10 years, five to 10 years, actuator development like the mechatronic development has improved massively like we have literally a new generation of motors now available, which was not in that form and price tag available 10 years ago. That involves the controller for that that involves the computational units to have a model based control that involves also sensors and sensing principles and it involves battery so power supply. And on all of these development lines, there has been a lot of advances have been made. And that combines into many people now saying I want to have my legate robot directly controlled and then very nice examples out there. And they work wonderfully and including my group, we were also participating on the development, the solo robot, which is an open source project out there. And we're using that and we will be using these kind of designs in future versions of birdbottles as well. But that's basically there's this one group of robotic leg designers that say we want to have it as we want to be able fully control. We do not want to rely on something which is not a point food because something which is not a point food is hard to model if you have an actual foot, a second-ended foot. And you have a changing center of pressure that's nothing so simple. And if I have springs in my system and maybe dampers, physical dampers in my system, then again I need to model those ones, they can react on their own, they have their own dynamics. And so they complicate my model based approaches. You need to have dedicated ways of controlling the more bi-inspired robots. And if you look at how bi-inspired robots are being controlled, like literally what kind of controllers are implemented, they are quite different. Do you try to solve a technical problem as an engineer? And you want to have a robot, a legate robot going up the stairs. You want to have it opening doors, carrying your package delivering. That's an engineering task which is wonderfully solved now with these robots. Or do we want to try to find a model and explanation, a proof of concept of what we observe in nature and can we actually have this proof of concept in a hardware robot, which creates an often advanced bi-inspired robot. That's a good point. Maybe a sense of course the end-to-half-year question. The first one, based on that, when you look for the end-goal that we design, machine that maybe similar exactly what we see, for example, on the ground part here. Where do you see the advancement will be? For example, robotics if we speak, do you imagine it would be more interesting design, morphology or actuation or material. And also for Monika, what would be that if you really unveil the secret behind what is happening in this animals or the ground part, what would be that would be open possibilities for robotics if we have exactly what is we see in that ground bird. Where the advancement will look like? I think there's some important advances to be made in balance control. There's some interesting sensing mechanisms that we're investigating and there's that could be important for balance control, localized sensing in the spinal cord of birds, for example. But as well as that coordinating with the sensing mechanisms, I think the foot structure plays an important role in balance control as well and enabling effective motion into three dimensions. Birdbot is not currently able to do fully 3D turning locomotion because it doesn't have actuators that can abduct and add the leg. So the advances might come in determining how to effectively couple intrinsic mechanics and sensing for effective balance control for three-dimensional locomotion. Yeah, the feet, absolutely. Bird feet are some of the most fascinating things. If you really look at them, if you go out in the park or in the zoo and you just start looking at bird feet, you will be surprised that none is like the other. There's so many specializations, adaptations, the change in segment length, they have webbing, they have different, like the way they thin out towards the end. Just the number of toes, you see like the ratio of those ones for the entire bird. And we do know that birds do different things. Many of these smaller birds purge. We don't see an ostrich purging, but we see relatively large birds can also purge, they can sleep on trees. So they have a specialized function where we do know that they can fall asleep doing that. And that strongly suspect there is a lot of functionality in the feet which allows to have specialized tasks being taken over by foot and leg structures. That's a really good point. Yeah, maybe for Monica, there is any example of animal still for you mysterious and locomotion. It's still pretty not very well understood. Yeah, where do I start? We are really only scratching the surface of understanding locomotion at this point. We have implemented bird bot is able to do steady gates in a straight line on a treadmill. And it demonstrates some of the essential features that We've identified in animals, which is following a mass spring like dynamics and storing elastic energy in the leg during stance. But we still have very little understanding about how animals integrate mechanics and control for maneuverability in complex terrains. And we have lots of pieces of evidence at different levels of our understanding. We have information about the sensors. We have information about muscle tendon organs and bones and skeletal geometry. We have information about the brain and the spinal cord and its functions. But what we still don't understand is how these are put together effectively to allow agile movement and the control architecture that's really involved in that. We have many hints, but we don't really have a good understanding of that. And until we do, we're not going to be able to recreate the kind of agile, maneuverable locomotion through complex unstructured terrains that we see, that evidently we see in animals, but still are not quite achieving in freely moving robots. Great. Yeah. For Alexander, maybe what are maybe other left question, maybe not answered when it comes to legate robots community or robotics internally. Still, this is really a question for salt. Yeah. Still, we have to guess this question. Do you have any other question that you think still should be considered? And maybe trade-offs, sometimes in design, sometimes in the trade-off. When you speak about the design, the trade-off, something-- yes, still not solved with a question or touch, yeah. Yeah. Like Monica there, it's endless. Every time we try to solve something and we think we found some kind of an explanation, we typically end up with 10 more very fundamental questions, which is exciting. That's exactly what we're looking for. And clearly, also in legate robotics, this very much like in the questions for animals, this concerns the hierarchy and the scaling of control and how control interacts with the robot body. In robots, we have a somewhat limited number of sensors typically. So we have joint sensors and we have pressure sensors and torque sensors and maybe an accelerometer. And if you look at that number and if you look at how much data is flowing into your processor, it can be more-- it becomes every year, becomes more-- you get better sensor, a higher resolution, faster sampling rate and so on. And on the other side in animals, what we know is we have just like multitude more sensors, which is-- we know they're all over distributed. They're not as selectively placed as we have that in robots right now or in software robots and in set there. And we use these-- as animals, we use these sensors for lifelong learning, for adaptation. So they make us adaptive and they make us evolutionary successful. And that's something which we want to achieve at some point in robotics by dealing with that amount of information that is clearly a challenge. And also inserting the right sensing type, just having those sensors available, like large strain sensors and soft robotics or so, that's a big field for good reasons. But even if let's assume we would have an equivalent of the sensor range and types, what we see in animals, we wouldn't know immediately what to do with that, right? So you want to make a decision. You want to have the all-centrally recorded, which requires ginormous computational power. I'm clearly not when I'm walking around. I'm clearly not thinking about exactly how much force each of my muscles is receiving and trying to coordinate and all of the individual muscle fibers. We have strong hierarchies and the only command which I'm thinking about, I want to run faster, I want to go left and right. Maybe I tilt a little bit my upper body. And that's the level of control which I typically give, like phase information, very, very high level. And at the same time, I know that all of my body, like every sensor is firing, and we just far, far away from that in robotics, so we often, the model-based controllers nowadays, are using direct information from the sensors and they're saying, OK, I implemented somewhere. The hierarchies, like deciding hierarchies, deciding where and which type of sensors actually needed is very unclear. And that relates to also what Monika was saying, there are not novel unknown and uncharacterized sensors in animals existing. Like the sensor in accelerometer, maybe, in this final quarter of birds. And they might actually also help to solve things in robotics. And again, what kind of information comes out? How is it being processed? How is it being used? Is it integrated directly into brain functions or is it indirectly used? That's a very large field of research. Yeah. That's excellent. Maybe the elastic question, what makes you-- yeah, for which you can answer the question fulfilled, what are you doing? And also, aspiration, in your love time, to answer these questions or the passion for doing what are you doing. So what makes you fulfilled, and also, aspiration in what you're doing? I think I'm a fund of NPTELI, curious person, who likes to solve problems. So what drives me scientifically is some level of innate curiosity in wanting to discover how things work. But I do find it very fulfilling to use that information to collaborate with colleagues who can then use this to develop new technologies or to inform rehabilitation of mobility in humans or to inform prosthetic designs. So I think movement in animals is a fundamental biological function. And activity and movement is essential for our well-being and health. So I feel that developing those fundamental principles in that fundamental understanding has wide-reaching applications that can help the general public just by understanding the importance of their own motions and maintaining healthy movement. But also, lots of more applied fields who can take a more principled approach to developing tools to rehabilitate people. So I think that that's what gives me the fulfillment is to see the applications of the knowledge. Yes, so I-- especially this research project here, for example, on the Bredbot project was extremely interesting for me to observe-- for myself, to make a direct bridge between the observation in the animal and being able to replicate this so much that we could see the performance of the robot. And as we discussed before, this is clearly related, also, to evolution, to form and morphology, and therefore, to mechanisms. And to be able to decipher that, in a way, this was very fulfilling to give you one example. And it was very-- yeah, it was just that. And of course, what comes now is the ability to transfer a desk technology into supporting human life every day, human life, ideally. That's where the engineer in me says that that would fantastic how can we make that. We have tons of problems in our society. And I believe having legs on the ground and being able to assert forces and carry around tools and to do that, or directly integrate this into prosthesis and exoskeletons equally can make a big difference. And I'm looking forward. I'm really hoping this will be integrated, and this will make a difference. And if that's the case, I would be very happy. Wonderful. So I don't know if you have any further words, I would like to see for Colzzy. Any further words, like to see? One short point. So for the listness of your podcast, our cut design is free access. So you can go onto our web page, and you find a link to our admin server. So you can directly access the cut files. And so you can replicate what if you want, or just the leg. And you can integrate it into your system. And so please do so and let us know. I'll put the link just at-- That's also for you. That would be great. [MUSIC PLAYING]