WEBVTT 00:00:03.557 --> 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 Verscher and Tony Prescott. 00:00:19.097 --> 00:00:23.977 This is Tony Prescott for the Convergent Science Network podcast from the Barcelona 00:00:23.977 --> 00:00:26.917 Cognition, Brain and Technology Summer School 2012. 00:00:26.917 --> 00:00:34.237 2012, and I'm here with Frank Grasso from the Biomimetics and Cognitive Systems Laboratory. 00:00:34.857 --> 00:00:42.117 A cognitive robotics lab. I'm sorry. At Brooklyn College in City University, New York. That's right. 00:00:42.397 --> 00:00:47.997 So, Frank, I've heard you describe your research as being about crunchies and squishies. 00:00:48.077 --> 00:00:50.437 Ah, yeah. Can you explain what you mean by that? 00:00:50.737 --> 00:00:55.077 Well, that's actually not my own terminology. It's a classic terminology from 00:00:55.077 --> 00:01:00.897 the old old neuroethological and neurophysiological studies when people studied 00:01:00.897 --> 00:01:09.277 invertebrate systems as quote-unquote simple alternatives to mammals and higher organisms. 00:01:09.497 --> 00:01:14.997 There was this taxonomy that divided the invertebrate animals into crunchies, 00:01:15.017 --> 00:01:19.017 which have a hard exoskeleton like crabs and lobsters and so forth, 00:01:19.077 --> 00:01:22.577 and squishies, things like octopuses and snails and so forth, 00:01:22.577 --> 00:01:24.177 that when you step on them, they go squish. 00:01:26.437 --> 00:01:29.577 So given that there are so many species we could study in the world, 00:01:29.677 --> 00:01:33.197 and some of them are more closely related to us than others, 00:01:33.617 --> 00:01:38.157 why would you want to study something which was clearly very different from 00:01:38.157 --> 00:01:42.077 us, like a crunchy or a squishy? Precisely because it is so different. 00:01:42.517 --> 00:01:47.097 These animals are of comparable size to us, the ones that I've chosen to study at least. 00:01:47.137 --> 00:01:50.877 They operate in a world that has the same sort of physics that we have to deal 00:01:50.877 --> 00:01:53.737 with, we humans have to deal with, and mammals have had to deal with, 00:01:53.817 --> 00:01:56.977 but their bodies and their brains have completely different architectures. 00:01:57.157 --> 00:02:01.557 And so they have succeeded in being able to cope with this world, 00:02:01.657 --> 00:02:05.377 and we get the contrasts and appreciate, 00:02:05.497 --> 00:02:10.677 really, how the problems can be solved, not just by one system. 00:02:11.077 --> 00:02:16.057 So would it be fair to say that your interest is in understanding the design 00:02:16.057 --> 00:02:20.977 space for animals by looking at the difference between. 00:02:21.815 --> 00:02:25.935 These different kinds of animal designs. Yeah, I think that's fair. 00:02:26.195 --> 00:02:30.795 And I've been pulled in the direction of thinking more about bodies than I would 00:02:30.795 --> 00:02:34.715 like because I've been principally interested in brains, and the brain architecture 00:02:34.715 --> 00:02:38.395 is the thing that I've really been driven to. 00:02:38.435 --> 00:02:43.435 I was trained as a computational neuroscientist, and so I've been pulled that way. 00:02:43.515 --> 00:02:46.215 And it's been an area of tremendous fascination to be able 00:02:46.215 --> 00:02:48.915 to understand the biomechanics and and and all of 00:02:48.915 --> 00:02:53.535 that connects to embodiment and situatedness that i have come to embrace in 00:02:53.535 --> 00:03:00.075 thinking about uh how brains actually uh realize behavior and brains have evolved 00:03:00.075 --> 00:03:06.495 in bodies in environments and the particular uh any particular species that you look at. 00:03:07.255 --> 00:03:11.175 In order to understand the brain you have to understand those other parts too, 00:03:11.695 --> 00:03:13.355 That's the idea, absolutely. 00:03:13.755 --> 00:03:19.335 And so in your lab, you're studying intact animals from the point of view of 00:03:19.335 --> 00:03:22.315 their behaviour, as well as doing some neurophysiology? 00:03:22.815 --> 00:03:27.295 Well, we have done neurophysiology in the past and we hope to do it in the future, 00:03:27.395 --> 00:03:33.175 but it's fair to say that most of what we've done recently has been behavioural 00:03:33.175 --> 00:03:34.615 work with the animals, yeah. 00:03:34.735 --> 00:03:40.075 And the biorobotics part? So the overall principle that we've worked on here 00:03:40.075 --> 00:03:45.635 is this idea of implementing a robot as a way of testing hypotheses about how 00:03:45.635 --> 00:03:49.915 behavior is coordinated and controlled within the organism. 00:03:50.055 --> 00:03:57.095 So we build tanks where we can test lobster behavior and then we can test the 00:03:57.095 --> 00:04:01.435 robots that we build under identical conditions to those that we test the lobsters 00:04:01.435 --> 00:04:04.235 with and then that becomes a yardstick. 00:04:04.235 --> 00:04:08.175 The lobster's behavior becomes a yardstick for evaluating how successful or 00:04:08.175 --> 00:04:13.735 unsuccessful our ideas about how behavior is coordinated and controlled are. 00:04:13.955 --> 00:04:17.675 And similarly with the octopus, we have the animals under the same roof in the 00:04:17.675 --> 00:04:22.675 laboratory, so we can look at their behavior under that context and then build 00:04:22.675 --> 00:04:26.995 robots or simulations and use the performance literally as a yardstick. 00:04:27.255 --> 00:04:32.815 So it's very hard to build a robot lobster. I guess it's pretty near impossible, 00:04:32.975 --> 00:04:35.555 given current technology, to build a robot octopus. 00:04:36.135 --> 00:04:42.655 So in order to get useful data from your robots, what do you regard as a sufficiently 00:04:42.655 --> 00:04:45.215 biological-like robot model? 00:04:45.555 --> 00:04:49.875 So I think what you have to do is to take the top-down approach and think about 00:04:49.875 --> 00:04:52.015 the theory that you're implementing. 00:04:52.954 --> 00:04:59.114 And so if your listeners are thinking about a robot lobster or a robot octopus 00:04:59.114 --> 00:05:02.554 that looks like a real lobster or looks like a real octopus, 00:05:02.794 --> 00:05:03.854 they'll be sadly disappointed. 00:05:04.814 --> 00:05:10.634 But if you're thinking about what a brain needs to do, then you can do this 00:05:10.634 --> 00:05:15.014 thing that I called biomimetic scaling, where you make certain that the key 00:05:15.014 --> 00:05:19.254 features of the system you're interested in are reproduced by that robot body. 00:05:19.254 --> 00:05:23.734 So things like the temporal processing capabilities of a sensor, 00:05:23.934 --> 00:05:28.874 the speed at which the motors allow the body to translate through space, 00:05:29.094 --> 00:05:33.474 the spatial sampling properties of the sensors, 00:05:34.114 --> 00:05:37.114 the central architecture inside of the robot. 00:05:37.294 --> 00:05:41.874 All of those things can be matched to the key points that are suggested by the 00:05:41.874 --> 00:05:43.254 theory that you're testing. 00:05:43.814 --> 00:05:49.474 So one of the areas that you've been working on is how lobsters are able to 00:05:49.474 --> 00:05:55.394 detect a source of chemicals that's in the water, a sort of olfactory source. 00:05:55.854 --> 00:06:00.914 And can you give me a brief description of what it's like to be a lobster trying 00:06:00.914 --> 00:06:02.114 to search for these things? 00:06:02.194 --> 00:06:07.314 And what are the insights that you've got from building robot lobsters that 00:06:07.314 --> 00:06:10.034 you might not have had just from studying the animal? Yeah. 00:06:11.234 --> 00:06:17.574 Well, I think it's difficult to really know what an animal experiences. 00:06:17.594 --> 00:06:21.054 There's the old ethologist's idea of the umwelt. 00:06:21.294 --> 00:06:26.254 And we can kind of get there by saying, what senses does the animal have and 00:06:26.254 --> 00:06:31.234 what information comes in to that animal's brain that it's able to use? 00:06:31.394 --> 00:06:34.734 And that's pretty much where I feel comfortable in terms of talking about the 00:06:34.734 --> 00:06:37.214 experience of the animal. I wasn't really going for consciousness, actually. 00:06:37.234 --> 00:06:44.434 I was just saying, imagine I'm sitting in the water and I've got lobster-like senses. 00:06:44.554 --> 00:06:47.014 Exactly, exactly. What kind of things are happening on my senses? 00:06:47.134 --> 00:06:50.814 Just to talk a little bit about the consciousness level, you don't even have 00:06:50.814 --> 00:06:51.734 to jump to consciousness. 00:06:51.934 --> 00:06:56.374 I think even understanding the perceptions of an organism, those things that 00:06:56.374 --> 00:06:59.214 are constructed from the sensory 00:06:59.214 --> 00:07:06.314 inputs, requires a lot of inference and a lot of… of interpretation. 00:07:06.634 --> 00:07:13.054 So, but, um, uh, from with that caveat in mind, the key phrase is intermittency. 00:07:13.494 --> 00:07:15.874 Um, when you, uh, 00:07:16.405 --> 00:07:20.865 want to understand how to build the sensors to put into a robot lobster, 00:07:21.025 --> 00:07:24.885 you look at the nature of the signals that are coming into the brain. 00:07:25.625 --> 00:07:28.885 And it turns out, and folks who've studied moths. 00:07:29.905 --> 00:07:34.885 Before us knew about this problem of intermittency as well in the air, 00:07:34.985 --> 00:07:39.785 but intermittency is basically the period of time during which the sensor is 00:07:39.785 --> 00:07:44.345 active, it's on, It's detecting the thing of interest out there in the world. 00:07:44.545 --> 00:07:49.405 And the amazing thing that we found when we actually went out and measured the 00:07:49.405 --> 00:07:53.865 environment that the robots would be operating in or that the lobsters had operated 00:07:53.865 --> 00:07:59.105 in was that there were long periods of time where you would be in a place where 00:07:59.105 --> 00:08:01.165 a statistical average would say, 00:08:01.245 --> 00:08:07.405 yes, you must be inside of the region of space that it's influenced by the source of the chemical. 00:08:07.505 --> 00:08:10.345 But in fact, it would be long periods of time where there was no signal. 00:08:10.345 --> 00:08:11.825 That's what we call intermittency. 00:08:11.905 --> 00:08:15.945 So you could be right there in the middle of, imagine a smokestack, 00:08:15.985 --> 00:08:18.585 right, as a source of odor pumping out. 00:08:18.605 --> 00:08:24.605 You could be standing in what looks like the output from a smokestack and see 00:08:24.605 --> 00:08:30.505 absolutely no chemical signal there for two minutes at a time, one minute. it. 00:08:30.525 --> 00:08:34.225 And then you might be hit with a whole barrage of these little intermittent, 00:08:34.385 --> 00:08:37.445 uh, uh, pulses of, of odor. 00:08:37.525 --> 00:08:41.545 So it's a, it's this kind of staccato world that the animal lives in. 00:08:41.605 --> 00:08:48.705 And during the talk, I played these audio recordings that we had made from a turbulent odor plume. 00:08:49.205 --> 00:08:54.505 And, uh, if we do think about the perceptions of a lobster and maybe use an 00:08:54.505 --> 00:08:55.445 acoustic analogy, analogy. 00:08:56.705 --> 00:09:03.765 My auditory system gave me a kind of musical quality to certain positions in 00:09:03.765 --> 00:09:06.285 the plume and a kind of non-musical quality. 00:09:06.545 --> 00:09:11.245 And if you take enough samples that way, you could begin to get a sense that 00:09:11.245 --> 00:09:16.725 maybe different parts of space that are influenced by chemicals from a particular 00:09:16.725 --> 00:09:21.245 source have a quality that could be described in aggregate from that. 00:09:21.245 --> 00:09:25.265 And that's one of the things that we learned from studying just the sensory 00:09:25.265 --> 00:09:28.765 umwelt of the animal, by directly studying the chemical distributions. 00:09:29.365 --> 00:09:34.805 So we're talking about your typical lobster now, not just any special lobster. 00:09:34.965 --> 00:09:39.385 Are they chemical sensing specialists in some way? Oh, that's completely true. 00:09:39.905 --> 00:09:45.105 It depends upon who you read and what interpretations you place on the primary 00:09:45.105 --> 00:09:50.865 data, but 70% of the lobster brain is estimated to be associated with olfactory reprocessing. 00:09:51.704 --> 00:09:56.324 And olfactory processing is important to them because it's for finding food, 00:09:56.404 --> 00:09:58.264 presumably. Yeah, they use it to find food. 00:09:59.904 --> 00:10:04.624 Many arthropods use it to find mates. It's probable that they use it to find 00:10:04.624 --> 00:10:07.724 mates, although it hasn't been—lobsters use it to find mates, 00:10:07.764 --> 00:10:10.224 although it hasn't been definitively demonstrated. 00:10:10.544 --> 00:10:14.644 There's a divide between animals that have pheromones and animals that don't. 00:10:15.364 --> 00:10:19.844 And no one has been able to find a lobster pheromone. There have been a few claims and so forth. 00:10:20.644 --> 00:10:27.384 So it is some evidence in the literature from Yela Atema that they actually do find mates that way. 00:10:28.044 --> 00:10:33.624 They also can smell states of decomposition 00:10:33.624 --> 00:10:37.544 and they can reject food items that are in a too bad a state. 00:10:37.624 --> 00:10:41.584 Although it's kind of funny to think about what would be too bad for a lobster to take. 00:10:42.104 --> 00:10:45.004 So they use it for those sorts of things to get at the quality. 00:10:45.364 --> 00:10:49.884 So that was definitely what they'd be using it for. And the Navy was very interested 00:10:49.884 --> 00:10:52.784 in American lobsters because the U.S. 00:10:52.824 --> 00:10:57.624 Navy, because they were so good at being able to solve the spatial localization problem. 00:10:57.864 --> 00:11:03.744 And presumably that's because they are trying to forage using the olfactory sense. 00:11:04.184 --> 00:11:08.424 And it's more, I'm sorry, 70% of the brain is what we're talking about. out. 00:11:08.864 --> 00:11:16.384 And, um, they are capable of very fine olfactory discriminations and have very, 00:11:16.464 --> 00:11:18.544 very low detection thresholds. 00:11:18.544 --> 00:11:23.284 Um, one analogy that I heard is that you could take a lake, the size of Lake 00:11:23.284 --> 00:11:28.344 Champlain and put a teaspoon of rose water in it, stir it up completely. 00:11:28.544 --> 00:11:32.664 And then the, the, the olfactory system would be able to tell the difference 00:11:32.664 --> 00:11:36.364 between the lake pre and post rose water inclusion. 00:11:37.124 --> 00:11:44.464 Wow. So it needs so many neurons in order to be able to discriminate so many 00:11:44.464 --> 00:11:46.064 different types of chemicals. 00:11:46.964 --> 00:11:52.904 It also needs to be able to do that to give it this very high fidelity for very weak signals. 00:11:53.444 --> 00:11:56.584 And then it needs to do this integration over time. 00:11:58.304 --> 00:12:03.484 So you've started to break down some of those aspects of the problem and to 00:12:03.484 --> 00:12:04.744 understand the mechanisms. systems. 00:12:05.504 --> 00:12:09.764 Which is the one that you find most interesting? Well, the one that we focused 00:12:09.764 --> 00:12:12.624 on has been the timing, the one that relates to memory. 00:12:12.844 --> 00:12:18.404 Others have done the work that's associated with what I call the what question, what am I smelling? 00:12:18.924 --> 00:12:23.944 I was talking about the 70% of the brain, but the one that leads into memory 00:12:23.944 --> 00:12:27.524 is definitely the temporal processing part, and that's part of the reason why 00:12:27.524 --> 00:12:30.784 I claim to have interests in things cognitive, right? 00:12:30.784 --> 00:12:35.764 So the temporal processing bit is using quite a lot of that olfactory brain? 00:12:36.624 --> 00:12:40.364 Well, we don't know how much of the brain is being used because we've never 00:12:40.364 --> 00:12:44.724 been able to go in and visualize, say, with voltage-sensitive dyes during the 00:12:44.724 --> 00:12:47.404 brain of an animal during a tracking episode. 00:12:48.238 --> 00:12:52.138 But from the anatomy, which is a strong guidepost, 00:12:52.418 --> 00:12:57.738 you can imagine that what happens is the first volley of chemicals comes into 00:12:57.738 --> 00:12:59.478 the brain, a region called the 00:12:59.478 --> 00:13:03.398 olfactory lobe, and that does the initial sorting of the what question. 00:13:03.398 --> 00:13:06.878 And there's a much larger lobe, which is called the accessory lobe, 00:13:06.878 --> 00:13:13.858 where olfactory information is processed that's come from the – sorry, 00:13:13.938 --> 00:13:17.238 the accessory lobe is processed coming from the olfactory lobe. 00:13:17.338 --> 00:13:22.038 But also, it's a point of convergence of flow information that's also been processed. 00:13:22.178 --> 00:13:26.038 So it's a second-order neuropil, second-order, beautifully organized, 00:13:26.278 --> 00:13:31.878 cytotectonically organized structure that has a huge amount of processing. 00:13:31.878 --> 00:13:39.078 And it's likely that those memory relationships, in fact, are discriminated 00:13:39.078 --> 00:13:42.638 there, stored there for some period of time in reverberating circuits, 00:13:42.858 --> 00:13:44.438 just from looking at the architecture. 00:13:44.918 --> 00:13:48.058 So it's likely, although we don't have very much direct evidence. 00:13:48.418 --> 00:13:54.238 And what are the sensors like that the animal's using? Oh, the sensors are amazing. Amazing. 00:13:55.198 --> 00:14:02.578 You would really be surprised at the sophistication of the chemical and fluid 00:14:02.578 --> 00:14:05.418 mechanical sensors on a lobster. 00:14:05.638 --> 00:14:11.098 So if you look at a lobster, there are a number of appendages coming off of the front end of it. 00:14:11.158 --> 00:14:16.138 There were the large antennae, but those aren't chemosensory or mechanosensory 00:14:16.138 --> 00:14:17.718 in the fluid mechanical sense. 00:14:17.978 --> 00:14:21.598 There's the ability to sense flow. They're like parking curb detectors that, 00:14:21.618 --> 00:14:26.558 you know, maybe like rat's whiskers that they kind of bump into the walls with. 00:14:26.678 --> 00:14:31.158 But if you get down to the three centimeter long middle antennuals, 00:14:31.158 --> 00:14:36.978 and there are two pairs of them, there's one that has a quarter of a million 00:14:36.978 --> 00:14:41.358 primary chemoreceptors arranged along the length of that structure. 00:14:41.358 --> 00:14:45.658 The animal can move it through the water very quickly in sample space, 00:14:45.838 --> 00:14:51.618 or it can hold it stationary as it does when it's tracking and get a large volley of information. 00:14:51.778 --> 00:14:54.098 And that, that, um... 00:14:55.080 --> 00:14:59.980 Antenual has many segments, and each one of those segments is the home of about 00:14:59.980 --> 00:15:07.220 300 primary chemoreceptor neurons that do this decomposition and fragmentation 00:15:07.220 --> 00:15:08.800 of the chemical composition. 00:15:09.840 --> 00:15:16.980 Then along the sides of those are these nine odd different kinds of flow detectors. 00:15:16.980 --> 00:15:23.060 And so on that one structure, you have chemical information coming in in one 00:15:23.060 --> 00:15:27.120 physical and temporal context and mechanical information about the flow. 00:15:27.280 --> 00:15:30.780 And then those flow into the brain structures that I was talking about. 00:15:30.820 --> 00:15:33.400 The information flows in there that way. 00:15:33.540 --> 00:15:38.820 And so with two of these, the animal uses two of them to be able to tell the 00:15:38.820 --> 00:15:43.040 difference in the quality of chemicals on the left and right side as it's moving 00:15:43.040 --> 00:15:46.560 into a flow, as well as the intermittency that I was talking about. out. 00:15:46.900 --> 00:15:50.340 And these sensors all go through a process called adaptation. 00:15:50.740 --> 00:15:55.620 They actually adjust their firing rate with fatigue. So it's a dynamic system 00:15:55.620 --> 00:15:59.820 that can make for some temporal processing in the periphery from the very beginning 00:15:59.820 --> 00:16:02.000 that make the story richly complicated. 00:16:02.160 --> 00:16:06.500 And also probably a reason why the Navy was so interested in these animals, 00:16:06.620 --> 00:16:11.720 because they've tapped many levels of organization of the world with this elegant sensor. 00:16:12.200 --> 00:16:16.720 So I think despite objecting to my question, you have actually given me quite 00:16:16.720 --> 00:16:18.720 a good impression of what it's like to be a lobster. 00:16:19.960 --> 00:16:24.880 But you haven't addressed my second half of my question. What have we learned 00:16:24.880 --> 00:16:26.120 by building robot models? 00:16:26.540 --> 00:16:34.600 Right. So we learned some things about the nature of the problem that the animal has to solve. 00:16:35.100 --> 00:16:40.340 And fluid mechanicians would know about these sorts of things, 00:16:40.400 --> 00:16:44.940 but maybe not have put them together without thinking specifically about the lobster. 00:16:45.400 --> 00:16:52.400 And when you attempt to use the animal as a yardstick, you know what looks normal, 00:16:52.500 --> 00:16:53.500 what is normal behavior. 00:16:53.720 --> 00:16:58.440 So we built a robot lobster. We equipped it with sensors that would provide 00:16:58.440 --> 00:17:04.680 the information at the right rate and the right sensory interval and the right spatial organization. 00:17:04.980 --> 00:17:08.700 And we turned the lobster loose, the robot lobster loose in the same conditions 00:17:08.700 --> 00:17:14.020 that we turned to regular lobster loose, and the behavior of the agent was completely different. 00:17:14.280 --> 00:17:19.080 And so we were able to say, no, this mechanism that biologists have been saying 00:17:19.080 --> 00:17:23.960 for a hundred years is what animals like this are doing, can't work. 00:17:24.380 --> 00:17:27.280 All right, that isn't the fluid mechanics yet, that's the biology. 00:17:28.320 --> 00:17:30.180 And then we went on to test some others. 00:17:32.305 --> 00:17:38.225 We were able to define different regions at different distances from the source 00:17:38.225 --> 00:17:43.765 of the chemical where a very simple algorithm could be effective or completely ineffective. 00:17:44.105 --> 00:17:48.665 And we could do an information analysis of what the information coming into 00:17:48.665 --> 00:17:54.365 the robot had been as it made its decisions following our rules to say, in this region, 00:17:54.465 --> 00:17:58.265 there was absolutely no structure and no information for guidance if you did 00:17:58.265 --> 00:18:00.165 a purely chemical analysis. 00:18:00.165 --> 00:18:06.025 If you just used chemical information coming in, and that pointed us towards the need to use flow. 00:18:06.425 --> 00:18:11.565 Some flow information disambiguated it, and our performance measures became 00:18:11.565 --> 00:18:15.645 closer to those of what a lobster actually does. 00:18:16.905 --> 00:18:21.605 We're learning about different parts of the fluid mechanical regime that tell 00:18:21.605 --> 00:18:24.345 people who would like to build robots that can track chemicals. 00:18:24.805 --> 00:18:28.525 These are the problems that you have to face and solve. By the way, 00:18:28.565 --> 00:18:31.885 these are good ways of being able to tackle them and ways that won't work at all. 00:18:32.485 --> 00:18:37.205 On the other hand, we're learning about strategies that don't work for the biology. 00:18:37.905 --> 00:18:43.505 We're moving in both directions. So the robot allows you to look at explanations 00:18:43.505 --> 00:18:45.365 people have proposed and to 00:18:45.365 --> 00:18:49.605 give them a physical test in an environment very similar to the lobster. 00:18:49.885 --> 00:18:54.405 And if it doesn't work, you can be fairly sure it doesn't work in the real animal, 00:18:54.545 --> 00:18:59.345 even though your robot lobster is in many ways completely different from the real animal. 00:18:59.465 --> 00:19:04.105 But on the things that matter, some aspects of its sensing and processing, 00:19:04.425 --> 00:19:06.905 you've copied what you think is important. 00:19:07.225 --> 00:19:10.125 That's right. And that's that That idea of biomimetic scaling that I was talking 00:19:10.125 --> 00:19:15.905 about where you say this theory requires that the animal has this kind of temporal 00:19:15.905 --> 00:19:20.825 and spatial comparison on its sensors, this kind of central processing, and this kind of output. 00:19:21.664 --> 00:19:25.264 And it doesn't have to have sensors on its feet in order to be able to do it. 00:19:25.304 --> 00:19:28.284 It doesn't have to have legs in order to be able to test it. 00:19:28.564 --> 00:19:36.544 Scientific ideas are really ideas that kind of exist as abstractions of what we think is going on. 00:19:36.704 --> 00:19:40.284 And if you can slay that abstraction with evidence, you've actually managed 00:19:40.284 --> 00:19:46.284 to make the robot reproduce physically what the requirements of the theory are. 00:19:46.364 --> 00:19:51.484 And then it fails to reproduce the behavior of the animal. That's a very strong 00:19:51.484 --> 00:19:55.224 argument that the concept, the actual theory, isn't valid. 00:19:56.144 --> 00:20:01.664 Well, I think we started to understand some of the reasons for your work with crunchies. Yes. 00:20:01.884 --> 00:20:05.164 But let's talk about squishies. So they are fundamentally different. 00:20:05.964 --> 00:20:07.964 And your favorite one is octopus. 00:20:09.144 --> 00:20:10.464 What's special about octopus? 00:20:11.684 --> 00:20:15.584 Well, the special thing about the octopus is that it has no hard parts. 00:20:15.944 --> 00:20:18.684 If we look at the arthropods, the crunchies we were talking about, 00:20:18.724 --> 00:20:20.344 they have a hard exoskeleton. 00:20:20.584 --> 00:20:25.064 If we look at vertebrates like you and I, we have a hard endo, 00:20:25.064 --> 00:20:31.004 an internal skeleton that muscles act against in order to be able to produce action in the world. 00:20:31.104 --> 00:20:34.064 And octopuses don't have any hard parts. 00:20:34.124 --> 00:20:39.404 Well, they have two hard parts, but they really aren't valid for what the arms do. 00:20:39.544 --> 00:20:43.704 So an octopus can reach out, it can grab a jar, it can pick up the jar, 00:20:43.844 --> 00:20:49.224 it can rotate it very much like we would like it to be able to have a robot hand do. 00:20:49.224 --> 00:20:54.284 It can pick up a trash can or it could pick up a small glass or a pencil with 00:20:54.284 --> 00:20:56.744 the same basic mechanism, right? 00:20:56.964 --> 00:20:59.364 Wrapping the arm around and being able to do a grasping. 00:20:59.884 --> 00:21:02.564 There's some suckers that are involved that we'll probably come to. 00:21:03.424 --> 00:21:07.364 But the way that it does this is amazing if you think about the fact that it's 00:21:07.364 --> 00:21:09.024 basically an extension of the water. 00:21:09.384 --> 00:21:11.584 The arm of the octopus is just, 00:21:12.410 --> 00:21:17.150 an enclosed bag of water that reshapes itself as needed for the task. 00:21:17.370 --> 00:21:21.490 And without any hard parts, it does this by having muscle act against muscle, 00:21:21.630 --> 00:21:26.990 along with some connective tissue, which kind of guides the way that the muscles 00:21:26.990 --> 00:21:31.770 can work, but never provides the stiffness that our skeletons or the exoskeletons 00:21:31.770 --> 00:21:33.750 of crustaceans like lobsters produce. 00:21:34.130 --> 00:21:38.170 So you use the word hydrostatic to describe this kind of system. 00:21:38.250 --> 00:21:44.230 So what's the sort of internal constituency of an octopus arm that gives it this property. 00:21:44.610 --> 00:21:48.910 So it's always helpful to have a comparison. If you look at a starfish, 00:21:49.130 --> 00:21:51.730 a starfish has what's called tube feet. 00:21:51.870 --> 00:21:55.870 And it's kind of fun to watch starfish, especially if you have a lot of patients 00:21:55.870 --> 00:21:56.810 because they're very slow. 00:21:57.610 --> 00:22:04.450 They have hundreds of feet underneath their bodies that move forward and backwards, 00:22:04.590 --> 00:22:09.190 and And it can attach chemically to the substrate below and propel the animal 00:22:09.190 --> 00:22:11.690 along by coordinating hundreds of these little feet. 00:22:11.870 --> 00:22:17.010 The way that those feet move is by an inflation and deflation of bags, 00:22:17.210 --> 00:22:21.450 like a pneumatic mechanism that we use when we have jacks that lift up cars and so forth. 00:22:21.930 --> 00:22:25.190 That's a hydraulic mechanism where there's a net movement of fluid. 00:22:25.550 --> 00:22:31.050 For the octopus, there's a so-called muscular hydrostat system where instead 00:22:31.050 --> 00:22:34.210 of having a net movement of fluid, there's no movement of fluid. 00:22:34.570 --> 00:22:39.710 And when the muscle contracts and works against that essentially fluid compartment, 00:22:40.110 --> 00:22:44.950 it'll increase the stiffness of that compartment locally and reshape it. 00:22:45.030 --> 00:22:47.190 And it's that hydrostatic principle. 00:22:47.390 --> 00:22:51.950 And I should say, the fellow that pioneered this is Bill Keir, 1985. 00:22:52.330 --> 00:22:55.670 And we're just scratching the surface of what all of that means. 00:22:55.870 --> 00:22:59.410 But the concept is simple and clear, and you can do it in your own backyard 00:22:59.410 --> 00:23:01.070 yard with a water balloon. 00:23:01.250 --> 00:23:04.290 You take a water balloon and you compress it in one dimension. 00:23:04.390 --> 00:23:07.390 There's no change in the volume of water, but the thing reshapes itself. 00:23:08.230 --> 00:23:13.270 So the octopus is full of water balloons pressing against each other. That's right. 00:23:13.470 --> 00:23:17.430 And how on earth do you use something like that to open a jar? 00:23:19.090 --> 00:23:24.670 Well, the bottom line is that nobody really knows everything about them and 00:23:24.670 --> 00:23:29.670 no one has been able to build one that does exactly what the octopus does. 00:23:30.150 --> 00:23:33.870 And part of the reason is that the water balloon analogy is not perfect because 00:23:33.870 --> 00:23:36.910 it isn't a balloon full of water. 00:23:36.990 --> 00:23:40.350 It functions that way in terms of some of its mechanical properties, 00:23:40.450 --> 00:23:41.730 but it's actually a massive muscle. 00:23:42.460 --> 00:23:46.720 And the muscle can be modeled as water in terms of its density and properties. 00:23:47.180 --> 00:23:52.120 So what you do is you have many, many muscle fibers at different orientations 00:23:52.120 --> 00:23:56.380 that pull against one another and reshape the geometry locally, 00:23:56.640 --> 00:24:00.060 and also change the local pressure to help reshape the structure. 00:24:00.060 --> 00:24:04.800 So to wrap around a jar, you could imagine that, and in fact, 00:24:04.820 --> 00:24:08.320 it's probably true that this is what happens with the so-called longitudinal 00:24:08.320 --> 00:24:11.700 muscles, that if the muscles on one side of the arm contract, 00:24:12.120 --> 00:24:16.380 it's going to produce an expansion on the other side and produce a bend, right? 00:24:16.520 --> 00:24:20.260 And so that's the simple idea for being able to produce this. 00:24:20.400 --> 00:24:25.420 And if you imagine three of these subsets of muscles down the length of the 00:24:25.420 --> 00:24:29.340 arm, you You could have a contraction on the left side at the base of the arm, 00:24:29.420 --> 00:24:32.640 the right side at the middle of the arm, and the left side towards the end of 00:24:32.640 --> 00:24:34.700 the arm, and you could have three bends there on the arm. 00:24:34.840 --> 00:24:37.900 And you could multiply that process. Instead of having three regions, 00:24:38.000 --> 00:24:42.820 you could have 12 or 24 and just keep going with this local activation. 00:24:43.200 --> 00:24:47.400 And that's why these are called hyper-redundant systems, because where our arms 00:24:47.400 --> 00:24:49.720 have only three major joints, 00:24:49.920 --> 00:24:53.680 an octopus can produce as many joints as it wants down the length of its arm, 00:24:53.720 --> 00:24:57.620 but this simple mechanism, and put them at any orientation that it chooses to 00:24:57.620 --> 00:24:59.680 by activating different muscle groups. 00:25:00.080 --> 00:25:04.360 Right. So for robot engineers, the idea of having multiple joints that you can 00:25:04.360 --> 00:25:07.980 actuate creates quite a nightmarish problem of control. 00:25:08.400 --> 00:25:13.820 So are there some clues into how we might solve this from looking at the octopus nervous system? 00:25:14.100 --> 00:25:16.800 Well, the nervous system and the structure of the arm. 00:25:17.580 --> 00:25:24.680 What I just described with bags of water balloons is really an isometric shape, 00:25:24.760 --> 00:25:26.280 the same shape in all directions. 00:25:27.597 --> 00:25:30.577 If you look at the structure of the octopus arm, you'll see that the muscles 00:25:30.577 --> 00:25:36.417 are not arranged in this perfectly homogeneous direction, arrangement. 00:25:37.017 --> 00:25:41.677 They have a structure with preferred masses of muscle in particular directions, 00:25:41.817 --> 00:25:47.097 and more importantly, connective tissue, which prevents certain types of actions from happening. 00:25:47.277 --> 00:25:50.877 It's not stiff like a bone, but it's enough to keep the muscles from tugging. 00:25:50.977 --> 00:25:55.617 And I think the study of the biomechanics of the design there is an interesting 00:25:55.617 --> 00:25:57.597 way to be able to get at adaptation. 00:25:57.897 --> 00:26:02.017 So there are structural features that the nervous system must know about as 00:26:02.017 --> 00:26:04.937 it attempts to control these systems and can take advantage of. 00:26:05.077 --> 00:26:10.477 So the nightmare that the engineers face gets some kind of what they would call 00:26:10.477 --> 00:26:16.297 a dimension reduction by having these structures built into the arm that provide 00:26:16.297 --> 00:26:20.137 local stiffness that can channel the characteristic directions of movement. 00:26:20.837 --> 00:26:24.957 The central nervous system of the octopus is huge. 00:26:25.897 --> 00:26:31.897 Um, it's, uh, it's, uh, can get something like five times 10 to the eighth neurons 00:26:31.897 --> 00:26:35.297 in the central nervous system. That's about a half a million, half a billion neurons. 00:26:35.477 --> 00:26:37.937 There are billions of neurons in the entire organism. 00:26:38.257 --> 00:26:43.197 Um, I'm resisting the temptation to talk about some of those interesting specializations. 00:26:43.557 --> 00:26:48.117 Uh, but the, um, one simple thing that Benny Hockner, not simple, 00:26:48.157 --> 00:26:51.897 one brilliant idea that Benny Hockner and Tamar Flash came up with for being 00:26:51.897 --> 00:26:55.357 able to understand how an octopus arm might work. 00:26:55.457 --> 00:26:59.637 So they looked at the reaching reflex of the octopus and saw that if you were 00:26:59.637 --> 00:27:04.617 to put a propagating wave of stiffening moving from the base to the tip and 00:27:04.617 --> 00:27:06.297 point the arm simply at the beginning, 00:27:06.457 --> 00:27:09.917 you'd be able to reach to any, sorry, point the arm from the base, 00:27:10.037 --> 00:27:15.197 you'd be able to reach to any point in space that you wanted to with a relatively simple motor program. 00:27:15.697 --> 00:27:19.037 And that could reside in the arm rather than being in the brain. 00:27:19.597 --> 00:27:24.997 And the nervous system of the octopus has three-fifths of its neurons at the 00:27:24.997 --> 00:27:28.777 central nervous system, those half a billion neurons I was talking about, 00:27:28.897 --> 00:27:33.157 three-fifths of them are located outside of that large central brain of the animal. 00:27:33.257 --> 00:27:38.757 And they are motor neurons, sensory neurons that form a ring that connect the 00:27:38.757 --> 00:27:40.877 various eight arms of the animal. 00:27:41.057 --> 00:27:45.957 And that's an organizing principle as well, I think, that hasn't been proven, 00:27:45.997 --> 00:27:50.997 but looks likely to be a way of, again, doing this dimension reduction. 00:27:51.257 --> 00:27:55.117 Instead of having the central controller know everything, the brain know everything 00:27:55.117 --> 00:27:56.197 about the state of everything. 00:27:56.337 --> 00:28:01.997 It can be more of an executive where the various arms through this large plexus 00:28:01.997 --> 00:28:06.057 of neurons, the brachial plexus, can negotiate with one another. 00:28:07.199 --> 00:28:12.559 The current state of the world and the current state of desired action and resolve the things that way. 00:28:13.019 --> 00:28:17.399 And I think that probably the top level explanation about the nervous system 00:28:17.399 --> 00:28:22.839 is that the octopus is endowed at birth with a basic behavioral repertoire, 00:28:22.899 --> 00:28:25.199 which makes it competent to move around. 00:28:25.679 --> 00:28:30.159 And then when we see an octopus learn how to unscrew a jar, which I showed in 00:28:30.159 --> 00:28:36.639 my talk, what's happening is they have learning capabilities that can extend that repertoire. 00:28:37.199 --> 00:28:41.619 Uh, maybe creating new behaviors, but maybe just working with some basic primitives 00:28:41.619 --> 00:28:44.299 that work for a floppy, right? 00:28:44.359 --> 00:28:46.699 Muscular hydrostat arm and reassemble it. 00:28:46.759 --> 00:28:51.219 And the learning capabilities of octopuses in just about every dimension they've 00:28:51.219 --> 00:28:54.439 been tested, sensory motor learning, memory, visual discrimination, 00:28:54.859 --> 00:28:58.459 all of those categorization, all of those things show that the animal has exceptional 00:28:58.459 --> 00:29:00.479 learning capabilities in its nervous system. 00:29:00.619 --> 00:29:04.019 And I think that's part of the dimension reduction as well. 00:29:04.079 --> 00:29:09.599 A capacity to come up with a solution rather than being able to arrive at all 00:29:09.599 --> 00:29:11.059 of the solutions and choose amongst them. 00:29:11.259 --> 00:29:15.079 Is it almost the case that if you have this kind of a body, you have no choice 00:29:15.079 --> 00:29:19.539 but to be good at learning because there may not be some other way of controlling 00:29:19.539 --> 00:29:21.679 it than to learn from experience? 00:29:21.999 --> 00:29:26.799 Yeah, I think that's a fair explanation, but there's another explanation that 00:29:26.799 --> 00:29:29.359 might be out there and I don't know which one of them is true. 00:29:29.719 --> 00:29:33.099 One of the things that's most surprising to people about octopuses is that they 00:29:33.099 --> 00:29:37.919 live for only a year And during that time, they go through an exponential growth rate. 00:29:38.059 --> 00:29:43.059 There are Pacific giant octopuses that in one year can go from being a few milligrams 00:29:43.059 --> 00:29:46.239 as a single egg to 90 kilograms. 00:29:47.419 --> 00:29:50.979 That's a huge, huge change in body size. 00:29:51.199 --> 00:29:54.539 And if your body is going to change by several orders of magnitude over the 00:29:54.539 --> 00:29:59.559 course of your life, you also might need a learning mechanism just to be able to control that body. 00:29:59.819 --> 00:30:02.259 But there's nothing to say that it couldn't be both as well. 00:30:02.259 --> 00:30:08.719 For a long time, people wondered why the short-lived octopus had this tremendous learning capability. 00:30:08.979 --> 00:30:16.139 Because in a mammalian-centric perspective, long-lived organisms were the organisms 00:30:16.139 --> 00:30:17.539 that needed to remember things. 00:30:17.939 --> 00:30:22.759 And so, it was thought to be a property of long-lived animals. 00:30:22.939 --> 00:30:26.699 This goes back to my original points that we find the alternative solutions 00:30:26.699 --> 00:30:29.859 when we begin to look at the way that animals operate. 00:30:30.079 --> 00:30:35.019 Learning in these animals is astounding, and it's probably an ecological adaptation 00:30:35.019 --> 00:30:39.319 to a short life and maybe a complicated body that's hard to control. 00:30:40.199 --> 00:30:45.879 Why do they not live for longer? Is there some reason why they would have selected 00:30:45.879 --> 00:30:53.919 to have short lives? Um, the idea that's been out there that I don't find satisfactory, uh, but, 00:30:54.605 --> 00:30:58.065 is a satisfactory answer. I don't find it completely satisfactory, 00:30:58.345 --> 00:31:06.445 is that they basically have pushed a snail's body plan as far as they can. 00:31:06.525 --> 00:31:07.425 Their muscles were slow. 00:31:08.165 --> 00:31:13.665 They develop three, they evolve three hearts and their circulatory system is 00:31:13.665 --> 00:31:17.085 at the limit of its capability to keep the body perfused. 00:31:18.385 --> 00:31:24.245 The idea here is that they can't live very long because they've pushed their 00:31:24.245 --> 00:31:28.165 physiology to the limit to be able to compete with vertebrates. 00:31:29.285 --> 00:31:33.725 It's called the live-fast-die-young hypothesis of Packard. 00:31:34.405 --> 00:31:42.885 And we know that it's not metabolically necessary for them to die young because nautilus, 00:31:43.025 --> 00:31:48.125 chambered nautilus, which are a branch different from octopuses, 00:31:48.145 --> 00:31:52.885 cuttlefishes, and squids, the ones we call the colioids, can live for 20-30 years, 00:31:53.085 --> 00:31:55.125 maybe more in fact in the field. 00:31:55.685 --> 00:32:00.605 So the physiology of the organism is such that it can support that metabolically. 00:32:02.085 --> 00:32:06.045 But nautilus are disappearing from the Earth in part because they really can't 00:32:06.045 --> 00:32:08.525 compete with the world as it is now. 00:32:08.625 --> 00:32:10.865 Nautilus go back 500 million years. 00:32:12.825 --> 00:32:16.625 It's a question about octopus, one of many, where we don't yet have the full answer. 00:32:16.845 --> 00:32:22.825 But one of the things about these octopus arms I think that when they're taking 00:32:22.825 --> 00:32:24.885 the lid off a jar it's most noticeable. 00:32:25.065 --> 00:32:28.225 It's not just the arm it's also the suckers that are important. 00:32:29.045 --> 00:32:34.705 So what's special about a sucker and how does that work together with an arm 00:32:34.705 --> 00:32:37.325 to do these really complicated tasks? Yeah, um. 00:32:38.545 --> 00:32:44.785 Roger Hanlon, a colleague of mine, has coined the phrase that for an octopus, 00:32:45.185 --> 00:32:47.605 the arm is a device for deploying suckers. 00:32:48.805 --> 00:32:52.645 And there's a lot of truth to that. The structure, the biomechanical structure 00:32:52.645 --> 00:32:56.385 of the arm that I was telling you about has a lot of specializations in terms 00:32:56.385 --> 00:33:02.065 of those connective tissues to make that sucker-bearing surface special in terms 00:33:02.065 --> 00:33:03.965 of the control and in terms of the musculature. 00:33:03.965 --> 00:33:06.885 There's no reason a priori to 00:33:06.885 --> 00:33:09.625 not have suckers all around the arm but for 00:33:09.625 --> 00:33:12.745 some reason the animal has them all along the lower 00:33:12.745 --> 00:33:15.885 surface the one that faces the mouth and so 00:33:15.885 --> 00:33:20.345 the bends that you see have a kind of a bias to avoid the suckers the suckers 00:33:20.345 --> 00:33:28.585 are incredibly clever use of the physical media that the octopus lives in suckers 00:33:28.585 --> 00:33:33.225 can generate tremendous adhesive forces with relatively low energetic input. 00:33:33.345 --> 00:33:37.325 Something attached to that sucker is very difficult to dislodge. 00:33:37.425 --> 00:33:41.485 And if you've ever wrestled with an octopus, as I have been required to do on 00:33:41.485 --> 00:33:43.085 several occasions in my laboratory, 00:33:43.405 --> 00:33:50.285 you know that not only is the animal capable of deploying 300 suckers per arm 00:33:50.285 --> 00:33:52.745 to attach to whatever surface it wants to attach to, 00:33:52.865 --> 00:33:58.085 but a single sucker, once you pluck a few of a few of them off can be almost 00:33:58.085 --> 00:34:01.445 impossible to get off until you induce the animal to release. 00:34:01.725 --> 00:34:07.685 And that's because the suckers enclose a volume of water and then compress it 00:34:07.685 --> 00:34:10.965 to the point where they can actually rip apart the molecular forces that hold 00:34:10.965 --> 00:34:13.545 the water apart, hold the water together if they need to. 00:34:13.825 --> 00:34:17.905 So the adhesion force of one sucker is tremendous. 00:34:18.325 --> 00:34:21.845 And then the adhesion force of two becomes tremendous squared. 00:34:22.125 --> 00:34:25.665 And when you get up to 300, you have something that's immovable. 00:34:25.785 --> 00:34:31.865 Very sensible design for an animal that needs to grasp things and hold on to them. 00:34:32.005 --> 00:34:37.405 Prey items, or to avoid being eaten, right? It's a natural adaptation for strong attachment. 00:34:38.085 --> 00:34:41.165 The really unexpected and neat thing is their mobility. 00:34:41.725 --> 00:34:46.325 And I've referred to the suckers as the fingers of the octopus. 00:34:46.885 --> 00:34:50.465 Where the arms, as I told you, can reach out and hit any point in space and 00:34:50.465 --> 00:34:51.605 can reconfigure themselves, 00:34:51.945 --> 00:34:56.165 the suckers not only can provide tremendous forceful adhesion, 00:34:56.225 --> 00:35:02.265 but can in fact do fine manipulations of small objects as the animal needs to. 00:35:02.325 --> 00:35:06.185 And that's something that people don't expect when they hear about an octopus sucker. 00:35:06.265 --> 00:35:08.325 The squids and the cuttlefish... 00:35:09.225 --> 00:35:10.725 They're like the toy guns that 00:35:10.725 --> 00:35:13.545 you had when you were a kid. You shoot against the wall and it sticks. 00:35:14.565 --> 00:35:17.885 Octopus, once it's got something attached, has this ring of extrinsic muscles 00:35:17.885 --> 00:35:22.185 around each sucker that allows them to rotate whatever is held with that vigorous 00:35:22.185 --> 00:35:25.045 grasp and move it forward, backward, to the side. 00:35:25.305 --> 00:35:29.365 And one of the really neat things is they can pass things from sucker to sucker 00:35:29.365 --> 00:35:31.885 with great, what we call, inter-sucker coordination. ordination. 00:35:32.425 --> 00:35:36.565 And the animals form a conveyor belt of suckers moving items that they like 00:35:36.565 --> 00:35:40.725 towards the mouth and a conveyor belt moving things they don't like away from 00:35:40.725 --> 00:35:42.405 the mouth to shoo them away. 00:35:42.845 --> 00:35:48.525 What do we know about the sensors on the suckers? Oh, they are loaded with sensors. 00:35:48.885 --> 00:35:53.905 There's about 10 to the fourth chemoreceptors along the rim of the sucker and 00:35:53.905 --> 00:35:57.265 a comparable, although smaller, number of mechanosensors. 00:35:57.405 --> 00:36:00.565 So these are chemotactile structures that give the 00:36:00.565 --> 00:36:04.405 animal a sense of um of everything 00:36:04.405 --> 00:36:07.385 that it's in contact with it's almost like they're tasting 00:36:07.385 --> 00:36:11.645 objects with their suckers they certainly do and the thing i was going to say 00:36:11.645 --> 00:36:15.665 is if you imagine an octopus moving along the floor of the ocean it's tasting 00:36:15.665 --> 00:36:20.145 everywhere it goes uh you know if you're holding an octopus it's tasting you 00:36:20.145 --> 00:36:24.205 uh it knows quite a bit about you maybe more than you want it to know. 00:36:24.345 --> 00:36:25.565 That's an uncomfortable thought. 00:36:27.425 --> 00:36:31.605 Yes. But that's the Umwelt question that we were talking about before. 00:36:31.845 --> 00:36:35.805 And if we limit ourselves to not go into wildest speculation, 00:36:36.185 --> 00:36:40.345 the world of an octopus is probably also dominated by chemotactile inputs. 00:36:40.605 --> 00:36:44.085 They have a tremendous, wonderful visual system. But remember, 00:36:44.225 --> 00:36:50.405 three-fifths of the nervous system feeds its way in through that brachial plexus. 00:36:50.465 --> 00:36:54.105 There's an awful lot of chemoreceptors out there, mechanoreceptors, 00:36:54.165 --> 00:36:57.785 like the lobster having correlated inputs for the same point in space. 00:36:58.485 --> 00:37:03.025 And a huge part of that sensory world is coming in through those suckers, 00:37:03.885 --> 00:37:06.705 and we just don't know enough about the central representations. 00:37:07.265 --> 00:37:10.145 You know, there's got to be some in there. We don't know enough. 00:37:10.645 --> 00:37:15.465 And so thinking about convergent evolution, it's interesting to compare the 00:37:15.465 --> 00:37:18.845 octopus arm with the vertebrate tongue, isn't it? 00:37:19.065 --> 00:37:26.105 Oh yeah, yeah. And that, again, is a hydrostatic system with huge numbers of chemoreceptors on it. 00:37:26.585 --> 00:37:28.685 So, I mean, how far can you push the analogy? 00:37:31.205 --> 00:37:36.845 I think that the analogy is very apt, except, of course, no, it's very apt. 00:37:38.505 --> 00:37:41.725 The mechanism is essentially the same, the musculohydrostatic principle. 00:37:42.285 --> 00:37:46.785 No hard parts in a tongue. And the thing that I really like to say that goes 00:37:46.785 --> 00:37:49.285 beyond that analogy is the communication bit. it. 00:37:49.685 --> 00:37:56.005 So a tongue, the tongue that I'm using right now, is capable of subtle reshaping 00:37:56.005 --> 00:38:02.325 a musket very quickly to make the words that people are hearing me say in this podcast clear. 00:38:02.405 --> 00:38:04.605 It shapes those things. You take out a person's tongue. 00:38:05.651 --> 00:38:08.891 They aren't able to actually speak and articulate and produce language. 00:38:09.111 --> 00:38:13.291 But the differences, we're in Barcelona, right? The differences in accent that 00:38:13.291 --> 00:38:17.731 I have as a former Bostonian, those are also communicated by the structure of 00:38:17.731 --> 00:38:19.931 the tongue, the differences in language, 00:38:20.151 --> 00:38:23.491 and they come out, if you've heard the local Catalonian speaking, 00:38:23.771 --> 00:38:24.991 they can come out quite quickly. 00:38:25.371 --> 00:38:31.271 So I talked about this important musculohydrostat aspect that really underscores 00:38:31.271 --> 00:38:35.191 the precision of control that you can have with this bag of muscle. 00:38:35.651 --> 00:38:42.931 The analogy with taste is also apt because you could think of the tongue as, 00:38:43.131 --> 00:38:45.471 if you forget about its linguistic capacity, 00:38:45.851 --> 00:38:49.671 as a way of deploying taste buds to make an evaluation about whether to accept 00:38:49.671 --> 00:38:52.591 or reject a particular food item. 00:38:52.771 --> 00:38:58.771 And there's a huge amount of knowledge about this across vertebrates and other animals as well. 00:38:59.071 --> 00:39:04.471 And elephant trunks are an exception to that. Right? 00:39:04.591 --> 00:39:08.131 Elephant trunks aren't loaded with chemoreceptors. They're that pure muscular 00:39:08.131 --> 00:39:10.491 sort of thing that can do heavy lifting. 00:39:12.391 --> 00:39:17.131 I was going to say that the human tongue has those chemosensors, 00:39:17.131 --> 00:39:20.831 and it's restricted to being, you know, more or less inside the mouth, 00:39:20.891 --> 00:39:25.911 whereas the octopus arm reaches out into space and explores the world and so forth. 00:39:26.491 --> 00:39:31.431 But there are chameleons which have this tongue which can strike food items 00:39:31.431 --> 00:39:32.951 at a very great distance. 00:39:33.731 --> 00:39:38.611 I don't want to quote the number, but it's longer than their body length. 00:39:38.751 --> 00:39:41.211 The tongue can extend out using a musculoskeletal hydrostat. 00:39:41.391 --> 00:39:45.891 And that would be somewhat like a squid, which also uses these as attack mechanisms. 00:39:45.991 --> 00:39:47.631 Exactly right. Throws them at prey. 00:39:47.871 --> 00:39:53.031 The squid tentacle strike. That one I know the exact number for. You can get 70% strain. 00:39:53.271 --> 00:39:58.111 That's a 70% increase in the length of the appendage in 500 milliseconds. 00:39:58.531 --> 00:40:04.371 But the chameleon puts my beautiful squid to shame in terms of its strain. 00:40:04.491 --> 00:40:07.951 Its strain is over 100%, which is pretty remarkable. 00:40:08.551 --> 00:40:12.251 100% increase, more than 100% increase in the length of the appendage. 00:40:12.491 --> 00:40:18.731 So these hydrostat systems are really very different from conventional actuator 00:40:18.731 --> 00:40:21.411 systems that people are using in robotics. Yes. 00:40:21.591 --> 00:40:27.051 Now, how far are we from building a hydrostat that might actually be usefully 00:40:27.051 --> 00:40:31.091 deployed as part of a robot? Yeah, um... 00:40:32.568 --> 00:40:39.048 Nobody has actually done one yet. It seems to me that it should be a relatively 00:40:39.048 --> 00:40:40.828 straightforward procedure. 00:40:41.648 --> 00:40:46.268 But I've seen octopus-like arms that are robotic. 00:40:46.348 --> 00:40:50.028 Yes. These will be hydraulic or driven in some more conventional way. 00:40:50.088 --> 00:40:54.388 Right. The Octarm project that I worked on had a pneumatic control. 00:40:54.608 --> 00:40:58.688 And that's a little bit more like the starfish feet I was telling you about 00:40:58.688 --> 00:41:02.948 where you could inflate or deflate, add fluid and reshape the structure that way. 00:41:03.088 --> 00:41:07.808 It was a beautiful system in terms of being able to produce elegant curves and 00:41:07.808 --> 00:41:11.548 captured a lot of the properties of elephant's trunks and octopus arms, 00:41:11.648 --> 00:41:16.088 but it wasn't literally a muscular hydrostat system because of the net fluid flow. 00:41:16.468 --> 00:41:19.528 And there's some wonderful work that's being done in Italy right now, 00:41:19.668 --> 00:41:24.568 modeling the octopus arm and that's tendon actuated. 00:41:24.788 --> 00:41:28.068 And they get wonderful, beautiful control off of that. 00:41:28.068 --> 00:41:33.188 But again, it isn't the real muscular hydrostat principle where there's no net 00:41:33.188 --> 00:41:37.228 movement of fluid and contraction in one dimension produces a controlled expansion 00:41:37.228 --> 00:41:41.748 in another direction to reshape and bend and adjust the shape of the entity. 00:41:42.228 --> 00:41:47.028 We don't yet know what the advantages might be of doing it in that way, or have you some ideas? 00:41:49.728 --> 00:41:54.308 Advantages of this are that hyper-redundancy. So for example, 00:41:54.388 --> 00:41:59.928 if you wanted to rescue someone who was buried under a building, 00:42:00.408 --> 00:42:06.508 an arm that can reshape itself with as many bends as you want could easily be 00:42:06.508 --> 00:42:08.528 able to navigate moving through that space. 00:42:08.628 --> 00:42:11.168 We have snake robots that do that sort of thing now. 00:42:11.328 --> 00:42:14.588 But this would be something that could support itself and attach itself to various 00:42:14.588 --> 00:42:19.368 points and be able to reshape itself, maybe get food or communication down to 00:42:19.368 --> 00:42:20.788 someone who's buried in some rubble. 00:42:20.788 --> 00:42:24.008 That's a dramatic example but in general operation 00:42:24.008 --> 00:42:27.508 in cluttered environments is a really good way to 00:42:27.508 --> 00:42:33.508 use these hyper redundant systems so shape-shifting robots is as well we might 00:42:33.508 --> 00:42:38.228 be in the future with this yes uh although i find it hard to imagine the terminator 00:42:38.228 --> 00:42:43.308 3 robot actually being realized with muscular hydrostats that's probably a good 00:42:43.308 --> 00:42:45.788 thing thanks very much frank You're welcome. 00:43:15.920 --> 00:43:24.080 Music.