WEBVTT 00:00:03.597 --> 00:00:10.517 This is the Convergent Science Network podcast. Leading researchers in the domain 00:00:10.517 --> 00:00:16.797 of neuroscience, brain theory and technology are interviewed by Paul Verschoor and Tony Prescott. 00:00:19.737 --> 00:00:26.397 So this is Paul Verschoor with Maria Kruzma, one of the speakers in our BCBT summer school. 00:00:27.037 --> 00:00:30.037 And um you focus very 00:00:30.037 --> 00:00:32.777 much in your talk about building an 00:00:32.777 --> 00:00:36.077 artificial fish right that was very much the topic the topic 00:00:36.077 --> 00:00:43.537 of your of your talk and of a lot of your research is fish yes so um and you 00:00:43.537 --> 00:00:49.797 started out with a very fundamental question like what when we talk about biomimetics 00:00:49.797 --> 00:00:55.517 what do we really copy right that was your fundamental question you started out with So tell me, 00:00:55.517 --> 00:00:59.777 what are we essentially copying in biomimetics and are we really copying? 00:01:01.577 --> 00:01:07.617 I think we're copying, but we really don't always think of whether we are copying 00:01:07.617 --> 00:01:09.757 the wrong or right things. 00:01:10.057 --> 00:01:15.897 There is an approach in biomimetics that you can call the naive biomimetics approach. 00:01:16.157 --> 00:01:21.817 You just copy everything. I think like first propellers had feathers because 00:01:21.817 --> 00:01:25.497 people thought the feathers are very important things. 00:01:25.977 --> 00:01:28.357 You know, on every flying object 00:01:28.357 --> 00:01:31.537 you have feathers, so you want to do propellers, you have feathers too. 00:01:31.837 --> 00:01:37.117 But in the end they understood that the helicopters are hovering very well without feathers. 00:01:37.197 --> 00:01:41.457 From the point of view of aerodynamics, this doesn't really matter in this case. 00:01:41.457 --> 00:01:47.677 And I think if we look back to our own research now, 10 years ago, after 10 years, 00:01:48.037 --> 00:01:52.737 the projects that we are doing right now could look equally funny because of 00:01:52.737 --> 00:01:58.057 these wrong ways and wrong approaches we are taking and copying irrelevant things. 00:01:58.057 --> 00:02:01.317 Like first cars in history they 00:02:01.317 --> 00:02:04.797 had a place in front of the car for 00:02:04.797 --> 00:02:08.697 a horse and it was there for a long time because cars 00:02:08.697 --> 00:02:11.597 involved from the horse carriage before people 00:02:11.597 --> 00:02:15.817 understood that these places for a horse but they are completely unnecessary 00:02:15.817 --> 00:02:21.917 and redesign the cars and i suppose we do the same things and if we ask a question 00:02:21.917 --> 00:02:27.757 like what should we copy or what is relevant and then we maybe can avoid taking 00:02:27.757 --> 00:02:30.737 this wrong direction once in a while. Mm-hmm. 00:02:31.697 --> 00:02:32.857 So, um... 00:02:34.049 --> 00:02:40.129 So, but if you now look at the fish case as a specific target of your research, right? 00:02:40.209 --> 00:02:43.129 So, what makes fish so interesting? 00:02:44.389 --> 00:02:49.389 Well, naive biomimetic approach says you fish are excellent swimmers. 00:02:49.909 --> 00:02:51.889 Nothing swims as well as fish. 00:02:52.889 --> 00:02:57.649 So, from this point of view, it makes sense to copy fish, see how it works, 00:02:57.769 --> 00:03:02.749 and see how does it compare to a propelled vehicle. 00:03:02.749 --> 00:03:05.529 But where you come to 00:03:05.529 --> 00:03:09.289 difficulties i'm a little critical now about biomimetic approaches 00:03:09.289 --> 00:03:12.489 actually in my own research too i think that propellers 00:03:12.489 --> 00:03:19.189 are wonderful inventions you see nature has been working on methods to travel 00:03:19.189 --> 00:03:25.289 underwater for millions and billions of years it never she never came out up 00:03:25.289 --> 00:03:30.389 with a propeller but it's a wonderful simple device and very powerful So, 00:03:30.689 --> 00:03:34.929 but definitely if you want, if you look at the technology of propellers, 00:03:35.149 --> 00:03:39.689 what happens there for the last, I don't know, tens of years, 00:03:40.009 --> 00:03:42.309 all the development is already very incremental. 00:03:42.689 --> 00:03:46.749 And if you want to make a breakthrough to come up with something completely 00:03:46.749 --> 00:03:53.869 different, then you could look at undulating bodies like fish have to try to 00:03:53.869 --> 00:03:55.749 design something different. Mm-hmm. 00:03:55.969 --> 00:03:59.529 But now, if you talk about propellers, you might find them fantastic, 00:03:59.689 --> 00:04:03.929 but in some sense, fish seem to be more efficient swimmers than... 00:04:03.929 --> 00:04:06.109 They are more efficient. They are more energy efficient. 00:04:06.289 --> 00:04:09.549 That's one way why you want to look at fish. 00:04:10.029 --> 00:04:13.109 And another thing is they're quiet. Right. 00:04:13.740 --> 00:04:17.680 They don't leave the wake behind them. It means actually, if you look at the 00:04:17.680 --> 00:04:22.060 wake behind the propeller, this is all energy got wasted in water. 00:04:22.600 --> 00:04:28.140 So in that sense, I could argue that the very positive statement you made earlier 00:04:28.140 --> 00:04:32.000 about propellers is maybe a little bit biased, but that's the current state 00:04:32.000 --> 00:04:33.700 of the art. It's just the best we can do. 00:04:34.060 --> 00:04:37.120 It's the best we can do, but I think we're doing very well. 00:04:37.260 --> 00:04:42.500 We're researching on the fish robots, but they're not very efficient at the 00:04:42.500 --> 00:04:45.120 moment. And so, really, do you understand what is a mechanism? 00:04:45.540 --> 00:04:47.240 What makes her so efficient? 00:04:47.700 --> 00:04:52.460 We still don't understand how we should mimic this technology. 00:04:53.140 --> 00:04:59.520 One important aspect is that fish have distributed action. 00:04:59.840 --> 00:05:03.600 They have hundreds and hundreds of muscle fibers all over the body. 00:05:03.880 --> 00:05:07.840 And if you look at the current technology we have in hand, What is a reliable 00:05:07.840 --> 00:05:12.500 technology is really still the technology of rotating motors. 00:05:13.200 --> 00:05:16.840 Okay, we have other fancy things that I also used to work with, 00:05:16.900 --> 00:05:22.160 like electric, electroactive polymers and artificial muscles and nitinol or 00:05:22.160 --> 00:05:23.780 all these contracting actuators. 00:05:23.880 --> 00:05:27.100 But they have their own issues. This is not a mature technology. 00:05:29.300 --> 00:05:33.880 And now if you want to use a mature technology that we have, 00:05:33.880 --> 00:05:37.380 Now it's just DC motors and make a distributed actuation system. 00:05:37.860 --> 00:05:41.640 You end up with something very, very big and clumsy in the end. 00:05:41.700 --> 00:05:44.960 You end up with big and clumsy. It becomes energy inefficient. 00:05:45.180 --> 00:05:50.420 It becomes hard to control and really, really big. You don't make little agile, 00:05:50.560 --> 00:05:52.300 simple devices out of that. 00:05:54.500 --> 00:05:57.540 So we agree that in terms of our technology, 00:05:58.365 --> 00:06:01.445 best solution to the problem fish are still doing better okay 00:06:01.445 --> 00:06:04.485 so swimming fish do better fish do 00:06:04.485 --> 00:06:08.425 better so but then what i would like to know um 00:06:08.425 --> 00:06:13.785 what makes them better okay what are the tricks what are these principles that 00:06:13.785 --> 00:06:18.325 as a biomimetic engineer you might want to extract from fish to build better 00:06:18.325 --> 00:06:22.305 fish like machines right so what makes fish better swimmers every Every machine 00:06:22.305 --> 00:06:27.785 works by transmitting energy from their own body to the surrounding water. 00:06:28.905 --> 00:06:33.865 By transmitting the momentum of energy, they push themselves forward and move the water. 00:06:34.145 --> 00:06:38.185 The way propellers do it is by rotating the plates. 00:06:38.425 --> 00:06:44.085 The way fish do it is by undulating their bodies and creating a traveling wave. 00:06:44.365 --> 00:06:52.645 And then it passes the momentum of the water. and how do they do it so efficiently? 00:06:53.865 --> 00:06:57.725 We didn't know exactly. So there are, again, many aspects of it. 00:06:57.825 --> 00:07:00.465 One is the body, embodiment itself. 00:07:02.285 --> 00:07:07.445 Now when it comes to physics of compliant bodies and water interaction is a 00:07:07.445 --> 00:07:13.625 terribly, terribly difficult field to go into to understand how compliant bodies 00:07:13.625 --> 00:07:15.545 and water interact with each other. 00:07:15.545 --> 00:07:22.105 There is no mathematician who could sit down and write up the equations and 00:07:22.105 --> 00:07:24.245 say that this is optimal solution to the problem. 00:07:24.645 --> 00:07:28.925 Nobody can do that. So what we have is more like an experimental approach. 00:07:29.265 --> 00:07:35.745 We try to optimize surface area or the fins or we try to optimize the length 00:07:35.745 --> 00:07:37.785 of the body or whatever comes to your mind. 00:07:37.965 --> 00:07:42.265 And see that if we get more efficiency out of the fish. But now, 00:07:42.305 --> 00:07:47.685 one thing that's interesting about fish is that they actually have developed 00:07:47.685 --> 00:07:49.825 a pretty advanced sensing system. 00:07:50.821 --> 00:07:57.461 To be able to really measure in detail the properties of the medium around them, 00:07:57.581 --> 00:07:59.481 right? So these are these lateral line sensors. 00:07:59.881 --> 00:08:03.341 Lateral line sensor is a very specific sensor. 00:08:03.581 --> 00:08:06.961 In a way, it's very general because it's common for all fish, 00:08:07.141 --> 00:08:10.821 all fish have a lateral line. 00:08:11.001 --> 00:08:15.621 But on the other hand, it's only common for those creatures who live underwater. water. 00:08:16.041 --> 00:08:20.901 Nobody walking on the ground or flying in the air doesn't have a lateral line. 00:08:21.241 --> 00:08:28.001 And one problem with mimicking lateral line sensing is that it is not clear 00:08:28.001 --> 00:08:34.721 understanding among biologists either how this marvelous sensing organ actually works, 00:08:34.901 --> 00:08:41.761 in which cases is it active and in which cases it's passive, in which behaviors, 00:08:42.021 --> 00:08:47.961 and how this information is processed in the brain of the fish. 00:08:49.021 --> 00:08:53.121 So if you want to do a biomimetic approach of a lateral line, 00:08:53.221 --> 00:08:54.681 it's like shooting in a dark room. 00:08:55.281 --> 00:09:02.741 But here comes another aspect of biomimetics, which I would call inverse biomimetics. 00:09:03.901 --> 00:09:07.721 We think of biomimetics that we go from science to technology. 00:09:08.281 --> 00:09:12.561 Scientists find out something, they discover things. things, 00:09:12.621 --> 00:09:17.021 investigate things that exist that already they discover, they make a discovery, 00:09:17.221 --> 00:09:19.321 and then they pass over to us engineers. 00:09:19.881 --> 00:09:24.501 And the engineers say, uh-huh, here is a great phenomenon. How do I get some 00:09:24.501 --> 00:09:27.381 use out of it? And then they develop the technologies. 00:09:27.841 --> 00:09:32.881 So what biomimetics actually does is it turns discoveries into inventions. 00:09:34.041 --> 00:09:37.821 But if you did inverse biomimetics, you could do the other way around. 00:09:38.081 --> 00:09:40.621 You turn inventions into discoveries. 00:09:41.461 --> 00:09:47.541 And this is a little bit what we're doing with our fish lateral line experiments in philosophy project. 00:09:48.181 --> 00:09:51.781 It's very hard for a biologist to investigate the lateral line. 00:09:52.921 --> 00:09:57.221 It's a lot of work and I have a very deep respect to biologists who are working 00:09:57.221 --> 00:09:59.961 with it because it takes lots of, lots of, lots of patience. 00:09:59.961 --> 00:10:06.961 For example, if you want to understand what is the role of a certain modularity 00:10:06.961 --> 00:10:11.761 of lateral line sensing, as a sense pleasure or the sense flow in certain behaviors, 00:10:12.041 --> 00:10:17.601 it's very hard or even impossible to knock out some parts of the systems as 00:10:17.601 --> 00:10:20.321 a pharmaceutical or surgical or do these experiments. 00:10:20.321 --> 00:10:24.461 And sometimes they left only behavioral tools, which is very, 00:10:24.521 --> 00:10:27.101 very hard work for people. 00:10:27.361 --> 00:10:33.501 And here you could take a robot or a technology as a method for biologists. 00:10:34.441 --> 00:10:38.221 Say that I have now my artificial robot here, artificial fish, 00:10:38.401 --> 00:10:40.561 and it has an artificial lateral line. 00:10:40.721 --> 00:10:45.981 What if I knock out the left part of the artificial lateral line and see what happens? Mm-hmm. 00:10:46.590 --> 00:10:50.170 Does it give you anything as a biologist to think about? 00:10:50.970 --> 00:10:55.670 And if you do the inverse biomimetics in the right way, you could have a very 00:10:55.670 --> 00:10:58.870 big impact to what biologists can find out. 00:10:59.050 --> 00:11:03.090 Right. And it also means that you might have what you want to shoot for then 00:11:03.090 --> 00:11:05.850 is a continuous interaction between, 00:11:06.050 --> 00:11:12.070 let's say, hypotheses generated from the scientific domain validated in the 00:11:12.070 --> 00:11:15.330 engineering with predictions going back into the science. 00:11:15.410 --> 00:11:19.570 Exactly. To close the loop. Exactly. That's the role the biomedics can play 00:11:19.570 --> 00:11:22.070 here, which we clearly see also in your project. 00:11:22.310 --> 00:11:27.590 Yes. But now, to look at the lateral line, essentially, it's a strip of sort 00:11:27.590 --> 00:11:32.330 of hairs that runs along or hair-like elements that runs along the side of the fish's body. 00:11:34.010 --> 00:11:39.250 And it also has similar sensors on its head or on other parts of the body. 00:11:39.510 --> 00:11:44.070 And it's used to sense exactly what? What's the lateral line exactly sensing? 00:11:44.710 --> 00:11:48.690 Lateral line has two modalities. Actually, there are two ways of measuring flows. 00:11:49.050 --> 00:11:52.470 One is that you can measure flow velocity, and another thing, 00:11:52.510 --> 00:11:53.270 you can measure pressure. 00:11:54.110 --> 00:11:59.270 In very few conditions, when you have a laminar flow, they are very easily related 00:11:59.270 --> 00:12:02.570 to each other by Bernoulli's law when they are inversely proportional. 00:12:03.010 --> 00:12:07.390 But otherwise, if you are in a difficult hydrodynamic environment, 00:12:07.390 --> 00:12:12.330 Then how does this different modality, do these different modalities add up 00:12:12.330 --> 00:12:14.870 to each other is also a very complicated interplay. 00:12:15.870 --> 00:12:21.930 But fish is equipped with both of the systems. They have canal lateral lines 00:12:21.930 --> 00:12:23.510 and can measure the pressure difference. 00:12:24.350 --> 00:12:32.430 And it has a surface lateral line, what biologists call neuromasts, that measure the flow. 00:12:32.430 --> 00:12:38.030 These are just simply like little cantilevers, little beams that bend in a flow, 00:12:38.190 --> 00:12:42.830 and depending on the angle of bending, you can figure out how strong is the flow. 00:12:44.691 --> 00:12:49.031 So what happens next is very complicated and nobody exactly knows what happens 00:12:49.031 --> 00:12:54.391 because some of the processing, signal processing of the signals is done in 00:12:54.391 --> 00:12:56.871 a, so to speak, hardware already of the fish. 00:12:57.171 --> 00:13:03.851 Because hair cells have a different stiffness and different height and they 00:13:03.851 --> 00:13:09.331 probably also work as some sort of filters, low-pass, high-pass filters. 00:13:09.331 --> 00:13:14.331 And some of the, if you think of how something like a fish, who is a rather 00:13:14.331 --> 00:13:19.411 unsophisticated, uneducated creature, could do such a sophisticated signal processing. 00:13:19.651 --> 00:13:23.991 So one answer to that is that it probably does it already in embodiment. 00:13:24.211 --> 00:13:29.271 The sensors do some of the filtering and cleaning up the crap from all the noise 00:13:29.271 --> 00:13:32.811 from what they're getting in. And then it goes to prey. 00:13:33.571 --> 00:13:37.611 And then the fish makes a decision based on what it feels. Yeah, 00:13:37.611 --> 00:13:44.351 but the decision this brain is going to make is in this domain of what's called rheotaxis, right? 00:13:44.411 --> 00:13:48.931 It's like the response to a current, right? 00:13:49.151 --> 00:13:55.711 That's one thing that fish do is that they respond to a stimulus which is a current or flow. 00:13:56.731 --> 00:14:02.851 The simplest form of rheotaxis is when fish orient their nose against the flow. 00:14:03.031 --> 00:14:07.051 There are several reasons of doing that. One reason is that if you're a migrating 00:14:07.051 --> 00:14:12.631 fish, you have to go upstream to find your spawning place. 00:14:13.031 --> 00:14:15.931 Then you have to know the direction of upstream. 00:14:16.971 --> 00:14:20.371 And this is a classical example of a real Texas behavior. 00:14:20.671 --> 00:14:25.491 But you could also just try to be in sweet spots on a flow where you get a lot 00:14:25.491 --> 00:14:27.931 of orders and a lot of food flowing by. 00:14:28.151 --> 00:14:33.311 And then you just sit there and open your mouth and hoping for the fresh plankton of the day. 00:14:33.431 --> 00:14:39.551 Exactly. So what is interesting about that, right, is that these fish with the 00:14:39.551 --> 00:14:44.951 lateral line sensor can also live in a world where you have these dynamic objects, 00:14:45.071 --> 00:14:48.131 which are essentially the properties of the flow. 00:14:48.591 --> 00:14:53.591 Right, where you can say, okay, here I have a flow profile that I like, 00:14:53.731 --> 00:14:56.571 but here it's more like a flow obstacle that I should avoid. 00:14:56.571 --> 00:15:01.451 Avoid so it's as if you walk through a mist right where you have high density 00:15:01.451 --> 00:15:06.271 and low density spots in a field filled with mist so which you would navigate 00:15:06.271 --> 00:15:09.851 and you would go for the open spots because that's where you would have more visibility, 00:15:10.531 --> 00:15:13.411 it looks as if they have a very, 00:15:14.264 --> 00:15:17.524 Pretty complex representation of the dynamics of the medium that they're in. 00:15:17.824 --> 00:15:23.664 It's an interesting question. I don't know how much spatial memory the fish have. 00:15:23.824 --> 00:15:30.684 But if you think of salmon, for example, that does orderly attacks and combining 00:15:30.684 --> 00:15:33.244 it with information from the flow. 00:15:33.384 --> 00:15:38.704 And it finds its home river where it was born once and then goes upstream and 00:15:38.704 --> 00:15:44.384 migrates. So maybe it definitely fuses this also with other kind of information. 00:15:44.624 --> 00:15:49.304 They use information from the magnetic field and in the air and all sort of different. 00:15:49.504 --> 00:15:55.264 So I think all fish like other animals are patient optimizers. 00:15:55.464 --> 00:16:01.264 They're just trying to find them a likelihood, maximal likelihood from all the 00:16:01.264 --> 00:16:02.384 information they get in. 00:16:02.384 --> 00:16:06.184 But if you're, I don't know what's going on in fish's brain, 00:16:06.304 --> 00:16:10.184 but if you take it from an information theoretic point of view and you look 00:16:10.184 --> 00:16:12.744 at this flow maps that fish possibly could have, 00:16:12.904 --> 00:16:15.124 then you can think of that fish could 00:16:15.124 --> 00:16:19.904 build up a really complicated map of discriminate hydrodynamic events. 00:16:19.904 --> 00:16:23.464 Events tells that there is a certain hydrodynamic event 00:16:23.464 --> 00:16:26.524 in one spot and certain in the other and then 00:16:26.524 --> 00:16:29.384 you can think that it maybe adds up as kind of a 00:16:29.384 --> 00:16:32.164 robot environmental map as a 00:16:32.164 --> 00:16:36.684 topological map or a grid-based map or something you don't know what happens 00:16:36.684 --> 00:16:42.064 in its brain but this is something very new that roboticists don't do they don't 00:16:42.064 --> 00:16:47.104 build maps based on flow information but fish probably have this somehow incorporated 00:16:47.104 --> 00:16:50.184 into their environmental mapping. Right, yeah. 00:16:50.564 --> 00:16:53.204 That's very exciting. But then... 00:16:55.494 --> 00:17:01.114 In some sense, we cannot ignore this whole issue that what starts to become 00:17:01.114 --> 00:17:06.014 important now is also your relative size in terms of the medium you're in, right? 00:17:06.074 --> 00:17:08.494 Like, for instance, if you're a whale, even though they're not fish, 00:17:08.634 --> 00:17:11.134 but let's say you have that scale, or let's say you're a whale shark, 00:17:11.314 --> 00:17:15.854 so you're still a fish, but you're very big, then in some sense, 00:17:15.914 --> 00:17:18.594 a lot of this turbulence stuff, you just don't care about, right? 00:17:18.634 --> 00:17:21.154 Because you will overcome a lot of this. 00:17:21.494 --> 00:17:29.174 I think sharks care a lot. It has been shown that shark skin is actually a sensing organ. 00:17:29.674 --> 00:17:34.894 And what sharks probably do is what is called eddy odoratexis, 00:17:34.914 --> 00:17:40.954 that they get in the information of the smell and combine it with what they 00:17:40.954 --> 00:17:43.414 feel with the rattler line about the local flow. 00:17:44.294 --> 00:17:48.494 And then they make decisions where to go based on this information. 00:17:48.494 --> 00:17:55.574 If you think of sharks' ability to localize plot, 00:17:55.774 --> 00:18:02.454 even few drops of plot far, far away, they must have this ability very fastly 00:18:02.454 --> 00:18:07.234 to… So that means it's not then relevant for navigation itself, 00:18:07.654 --> 00:18:11.774 but what it's used for is for source localization. 00:18:12.314 --> 00:18:18.314 We didn't know whether it's also relevant for navigation itself or just source localization. 00:18:18.574 --> 00:18:22.134 But, well, it's kind of, I don't know whether it's a relevant question. 00:18:22.254 --> 00:18:25.194 Why do you have to navigate anyway in order to get somewhere, right? 00:18:25.194 --> 00:18:30.914 If you're a really small fish, right, then in some sense you want to be really 00:18:30.914 --> 00:18:35.094 clear or you want to understand where the turbulence is because you have to overcome, 00:18:35.374 --> 00:18:39.814 you have to generate a sort of propulsive force to overcome the counterforce 00:18:39.814 --> 00:18:41.174 you receive from your environment. 00:18:41.774 --> 00:18:46.674 So there, the scaling of yourself to the medium, also expressed in this Reynolds 00:18:46.674 --> 00:18:49.314 number, will become pretty important. 00:18:49.594 --> 00:18:53.474 Well, if you are a whale shark, that's something you don't have to worry about too much. 00:18:53.834 --> 00:18:56.994 Yeah, but this is what is called the negatory of taxis. 00:18:57.014 --> 00:19:01.494 So instead of going into the flow, you try to keep away from the stream because 00:19:01.494 --> 00:19:07.294 you're such a small fish, you're afraid of being blown away with a stream. 00:19:07.294 --> 00:19:11.794 So it's equally important whether you have a big Reynolds number or a small Reynolds number. 00:19:12.154 --> 00:19:16.834 Probably your equipment is different and your mechanisms are a little different. 00:19:17.374 --> 00:19:20.294 Okay, so that's a bit the question I want to get to. 00:19:20.614 --> 00:19:24.674 So we are scaling this relationship or we quantify this relationship with the 00:19:24.674 --> 00:19:25.794 so-called Reynolds number, right? 00:19:25.814 --> 00:19:28.974 It tells you something about this relation between, let's say, 00:19:29.034 --> 00:19:32.874 on the one of the forces that the medium exposes you to and those that you can 00:19:32.874 --> 00:19:34.074 generate yourself. Right. 00:19:34.916 --> 00:19:39.656 And so then the question is, if I'm operating at, let's say, 00:19:39.676 --> 00:19:44.396 low Reynolds numbers, are my principles of operation qualitatively different 00:19:44.396 --> 00:19:46.696 as when I'm operating on high Reynolds numbers? 00:19:47.556 --> 00:19:51.236 And where is the transition exactly? So can you say something about that? 00:19:51.296 --> 00:19:52.736 Is that understood in some way? 00:19:52.996 --> 00:19:57.436 No, I don't think if I ask the same thing from biologists, there is no kind 00:19:57.436 --> 00:20:01.016 of firm answer that, yeah, we established that already. 00:20:01.016 --> 00:20:10.116 Both small and big fish have a lateral line, so it must mean it's somehow important. 00:20:10.616 --> 00:20:13.796 If you're living in this medium, you have to sense this medium. 00:20:13.996 --> 00:20:20.036 But nobody seems to be investigating what is the correlation between the environment 00:20:20.036 --> 00:20:23.056 you're living in and the topology of your lateral line, for example. 00:20:23.676 --> 00:20:28.116 Why some of the fish have more lateral line sensor in the head and none of them in the body. 00:20:28.216 --> 00:20:32.316 And some fish have lots of sensors in the rear part of their body. 00:20:32.636 --> 00:20:38.576 And does it depend on environment or the way they're feeding or lifestyle and 00:20:38.576 --> 00:20:40.316 other aspects of the lifestyle? 00:20:40.616 --> 00:20:45.356 Nobody has a beautiful, simple law to answer for that. 00:20:45.596 --> 00:20:51.116 That's interesting. So, because what is interesting there is that it goes a 00:20:51.116 --> 00:20:54.156 bit back to this earlier point that although first you think, 00:20:54.236 --> 00:20:57.376 okay, fish swim, they swim in a turbulent medium, 00:20:57.616 --> 00:21:01.016 they have to sense this medium to optimize swimming, so they need lateral line. 00:21:01.236 --> 00:21:04.676 But then actually when you push it further, you see the lateral line might actually 00:21:04.676 --> 00:21:05.956 be used for something completely different. 00:21:06.586 --> 00:21:11.526 And that swimming comes almost for free for fish because, as you also mentioned in your talk, 00:21:11.906 --> 00:21:17.046 that fish show a lower metabolic rate in turbulent environments as opposed to 00:21:17.046 --> 00:21:19.966 non-turbulent environments, which seems a very surprising result, 00:21:20.246 --> 00:21:23.286 isn't it? In a periodic turbulent environment. Okay, so explain. 00:21:24.066 --> 00:21:28.906 Periodic turbulence is on moderate Reynolds numbers when you have like beautiful 00:21:28.906 --> 00:21:34.846 regular vortices normally, say, appear behind some object. 00:21:34.846 --> 00:21:40.626 Object and there's some evidence biology biologists suggesting that fish like 00:21:40.626 --> 00:21:44.646 to be in environments like that and also if you catch a fish from environment 00:21:44.646 --> 00:21:50.246 like that it's and investigates its metabolic rate it appears that fish are 00:21:50.246 --> 00:21:53.286 less tired and there is a very interesting. 00:21:54.266 --> 00:22:01.686 Experiment done in george lauder's lab in harvard with jimmy leo and his colleagues 00:22:01.686 --> 00:22:04.806 with a dead fish swimming upstream. 00:22:05.746 --> 00:22:09.806 So definitely their lateral line is not concerned at all because you're dead. 00:22:09.946 --> 00:22:11.326 Exactly. You can't sense anything. 00:22:11.506 --> 00:22:14.286 Even more, you can't even actuate yourself. 00:22:14.586 --> 00:22:19.806 Okay. So the only thing you're left with is your body embodiment which somehow 00:22:19.806 --> 00:22:23.166 has to create propulsion. 00:22:24.266 --> 00:22:31.266 And the mechanism you can speculate behind it is the same mechanism that pushes the sail forward. 00:22:31.926 --> 00:22:38.886 Just by taking advantage of the pressure differences in the vorticity. 00:22:39.826 --> 00:22:48.906 And what is the role of red or white in finding so sweet environments they like to be in is not known. 00:22:49.426 --> 00:22:53.686 If you ask biologists, they can't say you have a karmic gating. 00:22:53.686 --> 00:22:58.886 This is a phenomenon when you stay in regular turbulence and you flap your tail 00:22:58.886 --> 00:23:03.446 periodically with the same frequency of vortex shedding, whether this is a passive 00:23:03.446 --> 00:23:05.486 or an active behavior. Mm-hmm. 00:23:06.127 --> 00:23:13.087 So we don't know. But now this result of the dead fish swimming upstream was 00:23:13.087 --> 00:23:15.667 published a while back, no? It's not very recent. 00:23:15.987 --> 00:23:20.447 No, it's not very recent. I think, I'm not sure, but I think it's 2004 or something. 00:23:21.087 --> 00:23:23.867 So are you aware of anyone being able to replicate that? 00:23:24.767 --> 00:23:28.647 I have no idea anybody has replicated that so far. 00:23:28.807 --> 00:23:34.087 I haven't heard of anybody repeating this experiment. And I've had plans to 00:23:34.087 --> 00:23:38.747 repeat them with robotic fish in my laboratory, but I haven't got so far yet. 00:23:39.107 --> 00:23:46.667 But what we are speculating by looking at the results of our own research could 00:23:46.667 --> 00:23:53.627 be that what really matters is very delicate interplay between the turbulence 00:23:53.627 --> 00:23:55.727 and the properties of the body. 00:23:55.727 --> 00:23:59.467 So what really matters probably is that if you happen to be of the right size 00:23:59.467 --> 00:24:04.067 with the right floppiness, perhaps, and in the right spot, you know, 00:24:04.087 --> 00:24:04.987 that creates this phenomenon. 00:24:05.407 --> 00:24:09.587 Just you don't see very often dead fish swimming and this could be the reason 00:24:09.587 --> 00:24:11.087 why. Right. Yes, exactly. 00:24:11.427 --> 00:24:16.667 What is very powerful about this is that you really see that the morphology 00:24:16.667 --> 00:24:19.547 itself is actually solving a heart 00:24:19.547 --> 00:24:25.267 problem that we intuitively might have thought of as requiring a brain. 00:24:25.727 --> 00:24:30.987 So, that means that the passive dynamics of the body is structured in a way. 00:24:31.774 --> 00:24:36.994 That it can actually propel itself, given that the medium has certain properties. 00:24:37.314 --> 00:24:43.274 The same phenomenon has been demonstrated on the ground with passive walkers. 00:24:43.614 --> 00:24:47.694 You put a robot that doesn't have any sensing or has any actuation, 00:24:47.934 --> 00:24:50.714 you put it on a ramp, and it just keeps walking down. 00:24:51.194 --> 00:24:53.994 But it's the same thing there, that you have to have the 00:24:53.994 --> 00:24:56.734 right slope of the ramp in order to make it 00:24:56.734 --> 00:24:59.834 occur how often uh in in in 00:24:59.834 --> 00:25:02.874 your life are you walking down the slope and only 00:25:02.874 --> 00:25:07.054 down the slope so probably because you're not doing that and normally their 00:25:07.054 --> 00:25:11.554 ground is rough so this is why you're also developed actuation sensing and i 00:25:11.554 --> 00:25:15.754 think the same holds for fish but if you think of the ramp what is different 00:25:15.754 --> 00:25:20.074 in when you're fluid is that you're in a microgravity environment or zero gravity 00:25:20.074 --> 00:25:22.454 environment compared to what you what 00:25:22.514 --> 00:25:25.154 you're feeling in a ground. What you're feeling on a ground is a gravity. 00:25:25.574 --> 00:25:30.394 And in water, you almost don't sense gravity at all. And it kind of suggests 00:25:30.394 --> 00:25:32.434 your embodiment has to also be very different. 00:25:32.874 --> 00:25:37.014 And now you have a solid ground behind you. You're walking. 00:25:37.214 --> 00:25:42.614 You're a passive walker. But I think when you're interacting with the flow, the flow is compliant. 00:25:43.054 --> 00:25:51.294 So I think it's more like compares to walking on a trampoline rather than along the street. 00:25:51.294 --> 00:25:55.734 And this is why I think also it's harder to demonstrate underwater. 00:25:56.494 --> 00:26:02.734 Exactly but now so okay now what we have what you're also what we see now in 00:26:02.734 --> 00:26:04.934 these biomimetic principles of control, 00:26:05.534 --> 00:26:08.774 the body gives you already a lot of functionality 00:26:08.774 --> 00:26:12.694 some people would call this a morphological computation although I find this 00:26:12.694 --> 00:26:16.874 is a bit of difficult there's also a concept from embodied intelligence which 00:26:16.874 --> 00:26:21.254 actually is a little wider but says the same that you You outsource some of 00:26:21.254 --> 00:26:27.314 the control burden to your embodiment without being conscious about this. Right. So, but now…, 00:26:28.048 --> 00:26:32.908 I'm a fish, I'm alive, I have a brain, so I still want to control my swimming, right? 00:26:32.968 --> 00:26:36.388 I cannot just flop around in some vortex and see where I'm ending up. 00:26:36.708 --> 00:26:41.768 I want to no turn left, right, go up or down, catch prey, what have you. 00:26:42.188 --> 00:26:45.508 So I'm going to control that. So let's first look at propulsion. 00:26:46.068 --> 00:26:49.828 So what do we know really about, let's say, how I control my swim speed? 00:26:50.588 --> 00:26:56.728 That's very interesting research, which is done by John Long in Vassar College. 00:26:56.728 --> 00:27:04.748 Some years ago, and he has very elegant experiments demonstrating what are the 00:27:04.748 --> 00:27:08.268 first-order control parameters of the fish. 00:27:08.448 --> 00:27:15.548 So what fish seems to control is tailbeat frequency and stiffness and amplitude. 00:27:16.068 --> 00:27:21.828 And if you look further into that, swimming speed is a second-order control 00:27:21.828 --> 00:27:26.648 parameter. You control your swimming speed by controlling your stiffness, 00:27:26.948 --> 00:27:29.308 by controlling your tail peak frequency, for example. 00:27:30.268 --> 00:27:33.508 And what biologists have also demonstrated is there is a very beautiful and 00:27:33.508 --> 00:27:38.848 simple control law, which says that swimming speed and tail peak frequency are 00:27:38.848 --> 00:27:42.308 linearly related to each other. 00:27:42.448 --> 00:27:50.788 For example, different from a tail peak amplitude, which is a completely independent variable. 00:27:51.408 --> 00:27:57.188 But this seems strange. I mean, you would expect if I have a bigger fin, 00:27:57.368 --> 00:28:03.368 right, a bigger surface, a tail fin, then I would need less beats to generate 00:28:03.368 --> 00:28:05.248 the same propulsion. No? 00:28:06.508 --> 00:28:12.748 But now you seem to say it was independent of the size, but dependent on the frequency. 00:28:13.648 --> 00:28:17.828 It depends on the frequency linearly. What it does with the amplitude is not 00:28:17.828 --> 00:28:20.368 known. It's how does amplitude contribute. 00:28:20.808 --> 00:28:22.288 So it do vary their amplitude. 00:28:22.748 --> 00:28:29.008 But there is not like a beautiful, simple relationship between how they control 00:28:29.008 --> 00:28:33.468 their amplitude and how they propel themselves forward. 00:28:33.608 --> 00:28:37.248 Okay. So what do we know there in terms of fish behavior? Do they, 00:28:37.328 --> 00:28:42.228 in some continuous fashion, control their beat frequency or do they have preferred 00:28:42.228 --> 00:28:44.928 frequencies at which they beat their tail? 00:28:45.788 --> 00:28:50.068 You always have a preferred frequency. And the frequency is a resonance frequency. 00:28:50.428 --> 00:28:55.248 Because on resonance frequency, you're beating your tail with a maximal amplitude. 00:28:56.388 --> 00:29:01.948 And what fish could do in order to control their amplitude then is not directly 00:29:01.948 --> 00:29:05.668 necessarily to change the amplitude, but change the stiffness. 00:29:06.008 --> 00:29:09.588 Exactly. If you change the stiffness, you get a different resonance frequency. 00:29:09.868 --> 00:29:12.828 And by having different resonance frequency, you get different amplitude. 00:29:13.148 --> 00:29:16.768 So when you want to have high frequencies, you're probably making yourself stiffer. 00:29:17.048 --> 00:29:22.048 And having a stiffer body gives you higher amplitude and higher frequencies. 00:29:23.088 --> 00:29:28.188 This seems to be what fish are doing. Another interesting thing is that if fish 00:29:28.188 --> 00:29:33.748 are steadily swimming, and the steadily swimming is called the cruising speed 00:29:33.748 --> 00:29:36.828 of a fish, which is about one or two body lengths per second. 00:29:36.948 --> 00:29:40.228 So if you're a fish and you're swimming at a cruising speed, 00:29:40.368 --> 00:29:45.148 you can go on forever and forever and forever. You even don't get tired almost. 00:29:45.788 --> 00:29:51.488 And it's shown that fish use only very few muscles to do that. 00:29:51.488 --> 00:29:57.528 Most of its body is completely relaxed, and it uses a few of the anterior muscles 00:29:57.528 --> 00:30:03.508 to create motion and pass a traveling wave along the completely or almost completely 00:30:03.508 --> 00:30:06.068 passive tail. So this is why they don't get tired. 00:30:07.328 --> 00:30:11.208 They're so good energy optimizers. That's impressive. It's something like when 00:30:11.208 --> 00:30:16.768 you go and you run a marathon, you don't keep your legs stiff. 00:30:17.088 --> 00:30:21.028 You're very relaxed, aren't you? Just because you want to conserve energy. 00:30:21.488 --> 00:30:24.808 To have all these 42 kilometers going. But if you're a sprinter, 00:30:25.028 --> 00:30:26.408 you do exactly the opposite. 00:30:26.568 --> 00:30:32.188 Not exactly the opposite, but you use much more muscles in order to go very 00:30:32.188 --> 00:30:33.728 fast in a long period of time. 00:30:33.868 --> 00:30:37.028 Right, exactly. And fish do the same thing. They use their other muscles, 00:30:37.168 --> 00:30:39.868 white muscles, when they have to do something very fast. 00:30:40.888 --> 00:30:44.768 Saves their lives, get a prey. So this is for rapid turns, bursts, 00:30:45.228 --> 00:30:46.948 acceleration, decelerations. 00:30:47.328 --> 00:30:51.128 And they use it very seldom, only when they have to, 00:30:51.128 --> 00:30:54.128 you because like a sprinter you get terribly tired of 00:30:54.128 --> 00:30:57.088 it sure but now so i'm a fish i'm 00:30:57.088 --> 00:30:59.848 flapping my my tail fin but i probably have 00:30:59.848 --> 00:31:02.688 some oscillator sitting in my brain somewhere that's driving this 00:31:02.688 --> 00:31:10.368 right and but now this oscillator is driving my the tail the the tail but i'm 00:31:10.368 --> 00:31:15.408 i'm varying i'm varying the stiffness of my body in the meantime as well right 00:31:15.408 --> 00:31:20.248 so um so then the question becomes, 00:31:20.508 --> 00:31:25.828 should I control this oscillator that drives? 00:31:26.088 --> 00:31:32.968 How do I control the fin, the tail fin oscillator relative to the stiffness 00:31:32.968 --> 00:31:36.048 control I'm performing? What's the scaling there? 00:31:36.548 --> 00:31:39.468 Yeah, I can answer you. I have no idea. 00:31:39.808 --> 00:31:44.008 Okay. But you agree, no? It would be a pretty important relationship. 00:31:44.128 --> 00:31:46.648 Absolutely. To manage. Absolutely. Absolutely. 00:31:47.368 --> 00:31:53.668 Okay. Okay, because you could basically just break your muscles or the motor plant in any other way. 00:31:53.828 --> 00:31:58.528 Absolutely, but I have no idea why they choose to control one parameter or the other. 00:31:59.408 --> 00:32:06.428 So it's hard to ask fish, right? We can try, exactly. Okay, so now we know that... 00:32:07.802 --> 00:32:14.302 How we're swimming, so we flap our tail to the frequencies, we control stiffness, 00:32:14.922 --> 00:32:19.422 but then how does the sensing feed into that, right? 00:32:19.482 --> 00:32:25.182 So, in some sense, I could then argue, well, I don't need much sensing to achieve that, right? 00:32:25.182 --> 00:32:31.662 Well, if you have to control your swimming speed, so you probably, 00:32:32.142 --> 00:32:37.722 what you do is like when you walk, you're walking with respect to some global reference frame. 00:32:37.962 --> 00:32:44.302 So from this point of view, I don't know whether flow sensing is significant 00:32:44.302 --> 00:32:47.262 here or not, because flow is also flowing, right? 00:32:47.362 --> 00:32:51.222 Sure. So try to swim in a river and understand where you end up. 00:32:51.222 --> 00:32:56.522 I don't know whether you ever got out swimming from a boat, and you're there 00:32:56.522 --> 00:33:00.962 in a flow, and the boat is here, and you're swimming, and you kind of feel you're 00:33:00.962 --> 00:33:03.342 staying still, or you're swimming in one direction. 00:33:03.622 --> 00:33:06.182 And then you turn around and say, oh, where is the boat? 00:33:06.642 --> 00:33:10.282 Oh, it's far, far away somewhere already. And you couldn't understand you're 00:33:10.282 --> 00:33:12.742 being carried away with the flow. Right. 00:33:13.082 --> 00:33:16.482 And probably fish have the same problem. I think they have the same problem. 00:33:17.162 --> 00:33:27.042 Yes. What you feel in swimming, you can just extrapolate from when you're swimming, you're feeling track. 00:33:28.362 --> 00:33:33.062 But the funny thing with the track is that, I don't know whether you ever thought 00:33:33.062 --> 00:33:35.402 about it, but we humans don't have a track sensor. 00:33:36.882 --> 00:33:39.822 True. Right? And we don't have a speed sensor. 00:33:39.822 --> 00:33:42.522 Sensor so if you're sitting in a 00:33:42.522 --> 00:33:45.422 car and you're driving very very fast way 00:33:45.422 --> 00:33:48.202 too fast on a very good road which is sometimes too when 00:33:48.202 --> 00:33:51.122 you have a very good car uh you're not 00:33:51.122 --> 00:33:54.782 feeling anything you're not feeling the speed you could feel 00:33:54.782 --> 00:33:57.602 if you close your eyes you could feel that you're almost still but what you're 00:33:57.602 --> 00:34:00.722 feeling is acceleration try to drive a 00:34:00.722 --> 00:34:03.622 car on a bumpy road on bad road and you're starting 00:34:03.622 --> 00:34:06.362 to pump and your inner ear is going to react and you're 00:34:06.362 --> 00:34:10.062 feeling the speed suddenly so uh why 00:34:10.062 --> 00:34:12.942 do you know that you're going fast if you're going yourself not 00:34:12.942 --> 00:34:16.082 by the car is that you're getting tired and this 00:34:16.082 --> 00:34:21.822 is probably what fish is feeling when it swims against the stream either slowly 00:34:21.822 --> 00:34:28.522 or or fast it's getting tired and it seems that the cruising speed of one two 00:34:28.522 --> 00:34:33.862 body lengths is something that makes them least tired this is why they choose this sort of, 00:34:34.402 --> 00:34:38.282 But could you imagine that the fish also adjusts its body stiffness relative 00:34:38.282 --> 00:34:40.062 to the turbulence it's in? 00:34:40.841 --> 00:34:46.601 Absolutely, I think so. Right. Because that's what also conserves a lot of energy, right? Absolutely. 00:34:46.901 --> 00:34:49.181 To be more compliant with respect to the medium. Yeah. 00:34:50.121 --> 00:34:56.241 But the other thing that was interesting in your data analysis is that you show 00:34:56.241 --> 00:35:02.301 that in these relationships between, for instance, speed and tail fin beats, 00:35:02.561 --> 00:35:04.441 that that relationship is essentially linear. 00:35:05.481 --> 00:35:08.561 Do you think that's a coincidence? Is that just by accident, 00:35:08.661 --> 00:35:14.321 or is there some biomimetic trick behind this? 00:35:14.641 --> 00:35:21.481 I don't know. I'm almost fascinated about that. You know, there are 30,000 species of fish in the world. 00:35:21.701 --> 00:35:25.361 And as far as I know, people haven't found a contrary example. 00:35:25.741 --> 00:35:30.661 Why this law doesn't tell. So it's hard to believe for me, because I've always 00:35:30.661 --> 00:35:34.261 thought that such a general law only holds in physics. 00:35:34.801 --> 00:35:42.141 Because physics is physics. Like it's an unconscious world and it can't change in certain ways. 00:35:42.301 --> 00:35:47.721 But biology is normally, it's like, you know, biology is always imprecise and 00:35:47.721 --> 00:35:50.881 it's slippery and stinky and it can't be that beautiful. 00:35:51.381 --> 00:35:57.021 Exactly. But actually there are also other laws in biology, more or less, 00:35:57.041 --> 00:36:00.181 all statistical laws actually. In physics, you have precise laws. 00:36:00.281 --> 00:36:05.461 In biology, you have statistical laws. But within the reasonable statistical 00:36:05.461 --> 00:36:09.381 interval, these laws all hold. 00:36:09.601 --> 00:36:13.621 And why is this like this? I don't know, but I'm very happy for that. 00:36:13.961 --> 00:36:18.621 Because if I'm a control engineer, you know, there is nothing better for a control 00:36:18.621 --> 00:36:21.461 engineer than a linear control law. Exactly. 00:36:22.181 --> 00:36:26.281 Although maybe the precision you might find in physics is a bit overrated, right? 00:36:26.341 --> 00:36:29.101 If you go to… Oh, yeah, yeah. So if you go down to, let's say, 00:36:29.121 --> 00:36:32.221 the quantum level, then suddenly things become all probabilistic as well. 00:36:32.441 --> 00:36:37.281 Absolutely. You have measurement errors and all these sort of things. But now…, 00:36:37.735 --> 00:36:42.555 So now we've understood a lot about fish swimming, and subsequently you want 00:36:42.555 --> 00:36:43.815 to build your robot to swim. 00:36:44.115 --> 00:36:48.315 So how is that translation step? How do you do that? How do we go from this 00:36:48.315 --> 00:36:53.995 now, our understanding of fish swimming, to actually a fish robot in a tank swimming? 00:36:55.435 --> 00:36:58.755 We made many observations and tests from biology. 00:36:59.015 --> 00:37:03.155 As you pointed out now, the most recent one is a linear control law, 00:37:03.255 --> 00:37:07.735 which shows it also works for the fish. So we get a beautiful, 00:37:07.935 --> 00:37:10.775 easy way to control our fish swimming speed. 00:37:11.155 --> 00:37:17.095 And other things that we get from biology is the knowledge that if you swim 00:37:17.095 --> 00:37:21.335 on cruising speeds, you only use the anterior muscles of your body. 00:37:21.755 --> 00:37:25.755 What does it mean? It means that the rest of your body is just a passive carrier 00:37:25.755 --> 00:37:29.335 of the traveling wave that passes energy onto the water. 00:37:29.675 --> 00:37:33.415 And why it's very important from a technological point of view. 00:37:33.415 --> 00:37:39.475 As we started our discussion with distributed actuation, which is very hard 00:37:39.475 --> 00:37:43.475 to copy with the current technology we have, because we have rotating motors 00:37:43.475 --> 00:37:45.315 that are very big and clumsy. 00:37:46.255 --> 00:37:51.995 And what does it mean is that I can do a fish that has a single point of actuation only. 00:37:52.915 --> 00:37:56.395 And this gives me a very reductionist approach. 00:37:57.435 --> 00:38:01.855 So now, of course, everything depends on what my fitness function is. 00:38:01.855 --> 00:38:05.835 If I don't build a robot that has to fight for its survival, 00:38:06.035 --> 00:38:09.075 then I'm very happy with a single-point actuation. 00:38:09.375 --> 00:38:13.875 But if I want to do a robot that can turn around rapidly, that some researchers 00:38:13.875 --> 00:38:17.395 also in robotics are investigating, then probably this approach doesn't hold. 00:38:17.995 --> 00:38:22.875 It has its limits. But in this case, I'm very happy that I, from biologists, 00:38:23.215 --> 00:38:30.255 I derived a very reductionist approach that makes me able to develop simple robots. 00:38:30.255 --> 00:38:33.735 But simple means normally reliable and cheap. 00:38:33.915 --> 00:38:38.315 And these are two things that people appreciate in engineering. 00:38:39.395 --> 00:38:41.695 Yes. But now tell me how you... 00:38:42.272 --> 00:38:47.472 Practically built as a robot. What's the material you use for these fish, the robot fish? 00:38:47.752 --> 00:38:52.112 We use silicons. We use all sorts of silicons because we have a technology how 00:38:52.112 --> 00:38:53.352 we can vary the stiffness. 00:38:53.512 --> 00:39:01.052 And we can manufacture tails or whatever body parts you need with whatever elasticity 00:39:01.052 --> 00:39:03.352 and viscoelasticity you desire. 00:39:03.352 --> 00:39:11.892 And by that we can also find the right relationship between stiffness of the 00:39:11.892 --> 00:39:16.152 materials and the actuation momentums that we apply to the material. 00:39:16.392 --> 00:39:20.212 And then actuation, how is that done? Oh, it's done with a simple servo motor. 00:39:20.552 --> 00:39:24.512 There is fancier technologies, of course, with electroactive polymers, 00:39:24.672 --> 00:39:25.892 artificial muscles and everything. 00:39:26.112 --> 00:39:29.652 But if you wanted to do something cheap and reliable that people would buy in 00:39:29.652 --> 00:39:35.432 the end. So maybe it's a right thing to do is to go for the conventional technology first. 00:39:35.752 --> 00:39:43.612 So you have a name for this design methodology you pursued and you called it KISS. 00:39:43.832 --> 00:39:49.972 Well, KISS is a very general, a well-known design methodology in all engineering, 00:39:50.112 --> 00:39:52.112 which means keep it simple stupid. 00:39:52.112 --> 00:39:57.692 And I think most of people who come from engineering know what KISS means, 00:39:57.932 --> 00:40:02.132 especially software developments, where your software developers. 00:40:02.932 --> 00:40:08.632 The code gets out of your hands so fast that if you're not taking a KISS approach, 00:40:08.932 --> 00:40:10.052 you're doomed very soon. 00:40:10.052 --> 00:40:18.192 And, yeah, we take this approach because I believe that from all the things 00:40:18.192 --> 00:40:20.292 you can copy from biology, 00:40:20.632 --> 00:40:27.092 you have to copy only the most important, most relevant things to get some things 00:40:27.092 --> 00:40:28.332 that people can use later on. 00:40:28.904 --> 00:40:32.144 And which is easy to make simple, reliable, and cheap. 00:40:32.624 --> 00:40:37.464 But on the other hand, KISS methodology, I think, is also relevant in science. 00:40:38.184 --> 00:40:49.504 Because if you want to keep things simple, it helps you this kind of approach. 00:40:49.964 --> 00:40:55.044 Helps you to understand what is the importance of one or the other factor in 00:40:55.044 --> 00:40:56.664 the phenomenon that you are investigating. 00:40:56.664 --> 00:40:59.584 Complicating for example you take off the lateral line 00:40:59.584 --> 00:41:02.344 sensing and you ask what happens if i take it off 00:41:02.344 --> 00:41:05.984 i make my system very simple i take all sensors off what 00:41:05.984 --> 00:41:09.464 can i still do with the system without any sensors at all what can i do with 00:41:09.464 --> 00:41:13.864 a dead fish it's a very dead vicious simple right and then you establish well 00:41:13.864 --> 00:41:18.464 certain things i still can do so i don't need a complicated system and this 00:41:18.464 --> 00:41:23.784 gives me understanding what is actually the role of passive dynamics in a biological system. 00:41:24.064 --> 00:41:28.984 Okay, now I put actuation in and ask, what can I do if I have an actuator? 00:41:29.644 --> 00:41:34.124 And I establish what is this relevant for? And so on and so on. 00:41:34.344 --> 00:41:39.344 And by just, you know, step by step complicating my system and controlling all 00:41:39.344 --> 00:41:44.244 the parameters taking one or the other off, I can make generalizations. 00:41:45.304 --> 00:41:51.844 But now, in some sense, I could also argue that what you're proposing is is 00:41:51.844 --> 00:41:56.384 maybe also the opposite of KISS, right? 00:41:56.424 --> 00:41:59.124 Because if you look at the outcome of this, you're saying, look, 00:41:59.304 --> 00:42:01.124 if we want to understand fish swimming, 00:42:01.784 --> 00:42:05.504 we have to actually now look at a much larger picture because suddenly we have 00:42:05.504 --> 00:42:09.604 to include the morphology, we have to think about the actuation, 00:42:09.764 --> 00:42:12.944 the sensing, or the sensing might be partially removed but not completely. 00:42:13.844 --> 00:42:18.684 But, well, in a more traditional view, I could say, well, I have a more compartmentalized 00:42:18.684 --> 00:42:23.264 view on how a certain function is realized and have clear defined modules. 00:42:23.444 --> 00:42:28.244 They're well described with clear interfaces between them so I can understand what I'm doing. 00:42:28.304 --> 00:42:32.464 While you are turning all that upside down now because you're saying, well, actually, 00:42:33.184 --> 00:42:37.944 if the body itself can already swim and the actuation is actually modulating 00:42:37.944 --> 00:42:42.084 these properties of the body and then the brain must be sort of modulating again 00:42:42.084 --> 00:42:45.364 these properties of the actuation and the motor plant and so on. 00:42:45.524 --> 00:42:49.504 So how is it actually simple? It also can be a complexification. 00:42:50.484 --> 00:42:55.824 What you're describing here is a very fine, decent, respectable engineering method. 00:42:56.064 --> 00:43:01.704 You take some modules that exist already and you build a new equipment. That's fine with that. 00:43:02.284 --> 00:43:04.704 What I try to do is to create new models. 00:43:06.424 --> 00:43:10.324 In engineering, you don't have a module for an embodiment. 00:43:10.824 --> 00:43:15.084 So I'm not building a fish. I'm creating a method for building a fish. 00:43:16.004 --> 00:43:17.704 Yeah, but the point then is... 00:43:19.056 --> 00:43:23.516 You will have no more boxes if you want in that method, right? I invent new boxes. 00:43:24.076 --> 00:43:27.496 Yeah, but what's your box then? Do you have a box morphology? 00:43:27.796 --> 00:43:31.616 Yeah, I think I do have a box morphology. 00:43:31.836 --> 00:43:37.476 And even within this box, I know already how to, if you tell me that my body 00:43:37.476 --> 00:43:40.856 has to be with a certain viscoelasticity, I also have even, 00:43:41.116 --> 00:43:47.276 well, I'm going to have engineering methods how you fabricate materials like this. 00:43:47.516 --> 00:43:52.576 Okay. But then between the boxes, the interface between the boxes now also become constraints, right? 00:43:52.616 --> 00:43:57.476 It's not only control signals or sensory signals that are exchanged between 00:43:57.476 --> 00:44:00.076 these systems. Now these are also constraints. 00:44:00.416 --> 00:44:03.416 For instance, my body has a certain stiffness. 00:44:03.756 --> 00:44:06.756 Oh, yeah, but everybody has a certain stiffness. It's always been a constraint. 00:44:07.276 --> 00:44:12.556 No, but I mean, this is not anymore an explicit control signal you exchange, right? 00:44:12.616 --> 00:44:16.016 It becomes more a constraint on the control signals you might receive. 00:44:16.836 --> 00:44:21.576 Yeah, but I'm not very worried of that because every system has constraints. 00:44:22.016 --> 00:44:25.236 Every physical system has boundary conditions and constraints. 00:44:25.476 --> 00:44:29.096 So you just have to establish them and build your control on top of that. 00:44:29.816 --> 00:44:34.316 You're right, it can be complicated, especially when it comes to viscoelasticity. 00:44:34.476 --> 00:44:38.176 People talk about soft robotics, but they talk about elastic robots. 00:44:38.496 --> 00:44:41.176 They don't talk about viscoelastic robots. That's right. 00:44:41.376 --> 00:44:47.356 Why they don't talk about viscoelastic robots? Because viscoelasticity is a 00:44:47.356 --> 00:44:52.656 real pain for post-physicists and engineers to describe, to analyze, 00:44:53.036 --> 00:44:58.736 and we're not even talking here about control because I don't know any viscoelastic 00:44:58.736 --> 00:45:01.436 robots that people control. Right. 00:45:01.716 --> 00:45:09.816 So if we go from your biomimetics to, let's say, this KISS methodology you're proposing, 00:45:10.216 --> 00:45:16.076 do you see this as, let's say, a real paradigm shift in engineering or is this 00:45:16.076 --> 00:45:19.196 more of an elaboration of the standard practice? 00:45:22.576 --> 00:45:27.996 Well, I am ambitious enough to say it's a paradigm shift, of course, 00:45:28.096 --> 00:45:35.756 but on the other hand, I know people have tried it in different fields quite a lot. 00:45:36.436 --> 00:45:42.016 If you take a clever biomimetic approach, this is exactly the questions what you're asking. 00:45:42.336 --> 00:45:47.256 What features are relevant that I should copy? So people have done it maybe 00:45:47.256 --> 00:45:52.976 not so consciously, but there are many fine examples probably around already 00:45:52.976 --> 00:45:56.776 where people in one way or other are using the same approach. 00:45:57.056 --> 00:46:01.236 Right. But you are implying there's also a not so clever biomimetic approach. 00:46:01.876 --> 00:46:03.656 Yeah, yeah, yeah. 00:46:04.736 --> 00:46:11.016 Straightforward. So now tell me, what's your prediction with respect to the 00:46:11.016 --> 00:46:12.076 viability of this approach? 00:46:12.176 --> 00:46:16.356 When will we see the first artificial robot fish swim around the world, 00:46:16.476 --> 00:46:18.136 let's say, autonomously? 00:46:19.276 --> 00:46:23.856 You know, swim around the world autonomously actually doesn't depend exactly. 00:46:24.776 --> 00:46:29.976 I believe it doesn't depend on the technology I am developing. 00:46:30.056 --> 00:46:33.376 It depends on the technology of power sources. 00:46:35.076 --> 00:46:43.536 But we imagine we have it solved, we live off plankton, we have a plankton driven robot fish. 00:46:43.536 --> 00:46:47.836 Suppose we have a power source that lasts forever, so it's actually a very interesting 00:46:47.836 --> 00:46:52.276 test to do and it's a great idea actually to have a grand challenge of robotic 00:46:52.276 --> 00:46:54.136 fish swimming. Exactly, around the world. 00:46:54.476 --> 00:46:57.556 Instead of overseas or somewhere. Exactly. It's actually a great thing to do. 00:46:57.736 --> 00:47:03.436 Okay. I have to say is that our fish breaks down about every second day or so when we test it. 00:47:03.516 --> 00:47:08.256 So we're very far from it at the moment. No, but what are you expecting? 00:47:08.576 --> 00:47:11.556 So it's interesting, right? The power is still the biggest problem. 00:47:11.676 --> 00:47:15.276 That's also why robots won't conquer the world anytime soon. Right. 00:47:16.230 --> 00:47:22.350 But what's your expectation with respect to just the control of the swimming, for instance? 00:47:22.510 --> 00:47:27.310 So when do you think we can really have mature systems that will not propel 00:47:27.310 --> 00:47:31.430 themselves anymore with propellers as we know them, but really with fish-like 00:47:31.430 --> 00:47:33.110 bodies that are really swimming? 00:47:33.250 --> 00:47:39.230 Maybe I'm overconfident now here, but I think my beautiful fish would be definitely 00:47:39.230 --> 00:47:43.890 able to swim overseas unless it gets eaten or something. 00:47:44.090 --> 00:47:50.790 Okay. If I had to find manufacturing technology behind it, because what it really 00:47:50.790 --> 00:47:54.510 comes to is the next step in biomimetics. 00:47:54.550 --> 00:47:58.350 First you do your engineering, you develop new engineering approaches. 00:47:58.770 --> 00:48:05.270 Then you apply those engineering approaches. But then it comes to manufacturing, right? 00:48:05.390 --> 00:48:10.570 And manufacturing is a very important part of underwater robotics. 00:48:11.090 --> 00:48:15.530 So the biggest problem in underwater robotics… It's water. Sure. 00:48:15.990 --> 00:48:21.610 And the biggest problem is how to keep water out of the robot. 00:48:22.130 --> 00:48:28.330 And what we can demonstrate is with commercial underwater vehicles, 00:48:28.610 --> 00:48:30.590 where they manage to do it rather well. 00:48:31.330 --> 00:48:39.970 So I would say that if I had good manufacturing technology, my fish and power 00:48:39.970 --> 00:48:45.090 source would last forever. My fish would definitely be able to swim overseas. 00:48:45.450 --> 00:48:49.610 Another problem is how long does it take? Exactly. 00:48:50.470 --> 00:48:57.070 And another problem is where does it end up? Exactly. Like a letter in a bottle, no? Exactly. 00:48:57.970 --> 00:49:02.630 It's an expensive bottle. Right. But now, so to finish up, two last questions. 00:49:02.650 --> 00:49:08.030 So, Mariam, you're in this business for quite some time. you build up this your, 00:49:09.010 --> 00:49:12.770 biorobotics lab in Estonia so. 00:49:14.754 --> 00:49:19.734 What would be the one law of Maria that you would like us to adhere to in our 00:49:19.734 --> 00:49:21.514 biomimetic exploration of the universe? 00:49:22.494 --> 00:49:26.714 Maria's law, what is it? Oh, now you're going to ask something about, 00:49:26.854 --> 00:49:30.074 like, does God exist or something. This is a so general question. 00:49:30.314 --> 00:49:33.554 I don't have a single law. So I want to be very careful with it. 00:49:33.894 --> 00:49:39.174 All my gifts approach for the fish works fine. I'm very happy with that. 00:49:39.274 --> 00:49:41.214 But I didn't want to make grand claims. 00:49:41.614 --> 00:49:45.794 Something like, this is generalizable to everything. Take this method and apply 00:49:45.794 --> 00:49:49.534 to insects or apply them to other animals because then it comes, 00:49:49.654 --> 00:49:52.614 I think what happens and it becomes very general. 00:49:52.814 --> 00:50:00.334 And a very, very general law is applicable to everything, but it lacks details. 00:50:01.354 --> 00:50:06.014 So it's like I can give you a law, but this law is like, you know, 00:50:06.034 --> 00:50:10.334 I can also give you a law for playing piano. You know, the law for playing piano 00:50:10.334 --> 00:50:14.314 is that you have to press the right key with the right finger at the right time. 00:50:14.394 --> 00:50:20.994 And you always can play your pianos like that. But it doesn't help you really 00:50:20.994 --> 00:50:22.334 to play anything, right? 00:50:22.594 --> 00:50:33.034 So what I'm afraid of is to giving a similar very general suggestion like, you know, be happy. 00:50:34.634 --> 00:50:38.374 Don't worry, be happy or something like this. But since you're not a professional 00:50:38.374 --> 00:50:40.114 piano player, I can challenge you on this. 00:50:40.354 --> 00:50:47.334 So your KISS methodology could be an example of a principle or a law you would propose. 00:50:48.054 --> 00:50:52.894 Or what you tell me now, you're saying, look, all paradigms have limitations. 00:50:53.254 --> 00:50:57.674 This would be another potential law. So in that sense, it's not you have no laws, right? 00:50:57.874 --> 00:51:03.494 So we just have to agree on which one is the one today, right? 00:51:03.794 --> 00:51:06.574 So what would be the one today? day i'm still 00:51:06.574 --> 00:51:10.274 afraid of speaking trivialities if i say 00:51:10.274 --> 00:51:13.314 that yes hello you have to do in biomimetics is 00:51:13.314 --> 00:51:16.054 keep it seems so stupid it's just it's a little bit 00:51:16.054 --> 00:51:21.994 of a triviality but maybe one thing if i if i'm now thinking that i'm a biomimetic 00:51:21.994 --> 00:51:25.854 engineer and i want to have my problem which is something different maybe it's 00:51:25.854 --> 00:51:31.514 insect flight or something so maybe really what i would do is is like a physicist 00:51:31.514 --> 00:51:35.974 approach with taking there that we have a phenomenon And then we make a controlled experiment. 00:51:36.674 --> 00:51:42.074 We try to knock out some phenomena and investigate what is the impact to the 00:51:42.074 --> 00:51:47.434 other, to the performance of the system, and to apply this law within our box 00:51:47.434 --> 00:51:49.934 of biomimetics. But, of course... 00:51:51.033 --> 00:51:56.213 It's very hard to predict how scalable it is because the phenomena are dependent 00:51:56.213 --> 00:51:59.673 on each other and your measurements are imprecise. 00:51:59.773 --> 00:52:05.013 You're maybe not careful with some design parameters and so on. 00:52:05.173 --> 00:52:08.653 But also, on the other hand, still physicists have the same problem, 00:52:08.773 --> 00:52:09.953 right? Experimental physicists. 00:52:10.413 --> 00:52:14.893 They don't know how the parameters depend on each other and they're still doing it. 00:52:14.993 --> 00:52:16.713 That's why they do experiments, right? Yeah, exactly. Exactly. 00:52:16.733 --> 00:52:18.653 So I think this experimental methodology, 00:52:18.853 --> 00:52:21.853 if you want to have a key message, like what's the meaning of life or something, 00:52:22.033 --> 00:52:27.733 then it would be to have a controlled experiment like this to establish a relative 00:52:27.733 --> 00:52:34.813 importance of the phenomena to your system regarding your fitness function that you're aiming to. 00:52:35.353 --> 00:52:41.413 Okay. So keep on experimenting. Yeah. That's the law. Yeah, it turned out to be awfully trivial. 00:52:42.133 --> 00:52:47.153 Now we got the law. And then last one is a prediction. So if I go visit you 00:52:47.153 --> 00:52:50.973 five years from now, and I'm going to say, look, five years back, 00:52:51.073 --> 00:52:52.153 you made this one prediction. 00:52:52.373 --> 00:52:55.333 Today, I'm checking with you whether it came true. 00:52:55.733 --> 00:52:59.473 What is one prediction you would be willing to make today you're most enthusiastic 00:52:59.473 --> 00:53:02.553 about, you feel most committed to right now? 00:53:02.733 --> 00:53:07.793 Within biomimetics, you mean? Sure, within the envelope of our current discussion. 00:53:11.353 --> 00:53:17.373 I think there are some very, if you look at technology, there are some very 00:53:17.373 --> 00:53:19.913 general trends that are going on. 00:53:20.233 --> 00:53:24.653 One thing is that technology is robotic technology. 00:53:24.753 --> 00:53:29.133 If we talk about parametric robots, it's robots are moving closer to humans. 00:53:29.373 --> 00:53:34.133 So we're moving from industry to service, which means that they have to be safer. 00:53:34.473 --> 00:53:38.813 And one way to make them safer is to make them soft. That's just one solution 00:53:38.813 --> 00:53:42.433 to making some safeguards. There are others, of course, learning methods and everything. 00:53:42.933 --> 00:53:46.833 But one thing you can do is to make robots soft. 00:53:47.893 --> 00:53:51.393 And, of course, there are also trends of miniaturization. 00:53:52.816 --> 00:53:57.876 Happening at the same time and i'd 00:53:57.876 --> 00:54:01.136 say it it's maybe going to taught 00:54:01.136 --> 00:54:04.136 simplification in one in in some sense 00:54:04.136 --> 00:54:09.296 that said in traditional robotics where so very much emphasis have been on a 00:54:09.296 --> 00:54:15.176 control fine beautiful control of very complicated systems i think the prediction 00:54:15.176 --> 00:54:20.056 is really that people are going to explore rather how to keep the control minimal 00:54:20.056 --> 00:54:24.056 to make robots that are safe and robust. 00:54:26.176 --> 00:54:30.196 Independent on whether your control laws are exactly correct or not. 00:54:30.516 --> 00:54:32.896 Will we have those in five years? Is that what you're saying? 00:54:33.036 --> 00:54:37.556 Five years, we'll have more safe and robust robots? 00:54:37.876 --> 00:54:41.716 I think we definitely have more safe and robust robots. And I think also they're 00:54:41.716 --> 00:54:46.636 going to be compliant because what you can see is an increasing trend. 00:54:46.936 --> 00:54:50.196 More and more research papers are written on compliant robots. 00:54:50.196 --> 00:54:55.916 They are now more restricted to compliant robot arms for very obvious reasons, 00:54:56.136 --> 00:54:59.796 but I think it's also moving to other fields of the research. 00:55:00.016 --> 00:55:04.696 So I think the next step then with compliant robots, we have a good theory for 00:55:04.696 --> 00:55:07.996 making robots with a changing elasticity, changing stiffness. 00:55:08.236 --> 00:55:12.136 And we understand also a theoretical background of it quite well. 00:55:12.376 --> 00:55:16.796 But one thing that we have trouble understanding is viscoelasticity. 00:55:17.416 --> 00:55:25.256 Well, most of the bodies are viscoelastic, and they are set for particular reasons. 00:55:26.056 --> 00:55:29.236 It is that if you have a viscoelastic body, you dissipate energy. 00:55:29.496 --> 00:55:32.056 When you dissipate energy, you filter 00:55:32.056 --> 00:55:37.136 out disturbances, and filtering out disturbances is an important thing. 00:55:37.696 --> 00:55:42.256 Right. So I think where it is going in five years, you're going also to see 00:55:42.256 --> 00:55:46.856 research on this sort of embodied signal processing. 00:55:47.816 --> 00:55:51.136 Okay, that's a good one. All right, I come back to you and check. 00:55:51.616 --> 00:55:55.076 Well, yeah, it's a self-fulfilling prediction because I'm going to research 00:55:55.076 --> 00:55:58.956 on it anyway. So you come back in five years, I fulfilled my prediction. 00:55:59.796 --> 00:56:03.176 Okay, Marcia Kruzma, thank you very much for this conversation. Thank you. 00:56:05.536 --> 00:56:12.056 The csn podcast was produced by the convergent science network of biometrics 00:56:12.056 --> 00:56:18.616 and biohybrid systems a project funded by the european seventh research framework program, 00:56:19.796 --> 00:56:25.516 for more interviews recorded lectures or upcoming conferences in the field of 00:56:25.516 --> 00:56:31.576 biometrics and bio-hybrid systems, go to csnnetwork.eu. 00:56:31.600 --> 00:56:39.600 Music. 00:56:31.936 --> 00:56:33.756 And thank you for listening.