Search form

Donate Today


Fish Robots and Bio-inspired Engineering | Kristi Morgansen

See how engineers are learning from biological systems to solve problems beyond what our current technologies have been able to achieve.

Full talk transcript

Many animals such as the bat shown here have the ability to move and maneuver in tight and dynamic spaces that are unpredictable, in ways that engineered systems cannot  yet achieve even with a human operator. But if we did have systems that could achieve these capabilities, we would have a much better ability to address needs such as search and rescue, emergency response, collecting information for environmental monitoring — many things that we have goals of doing that we can't do as well as we'd like to right now.

One of the things that we work on in my research group is how to address these challenges using fundamental biological principles, in order to improve engineering design and operation. So what I'm going to talk about in the next little bit here is some other work that we do with underwater vehicles, specifically looking at this kind of application. Now, in particular, the vehicles that we work with are referred to as fish robots, and you might be thinking, those are kinda nifty, but really what are the serious applications for them? It's a perfectly reasonable assessment.

Consider the task of planetary scale environmental monitoring. So, we would like to be able to go out on a global scale and collect information to make better models of what's going on in the ocean, given information on temperature, salinity, chemical content in the ocean, species growth and decay, and then come up with better ways of knowing what's going to happen in terms of seismic activity, species tracking, weather patterns and transportation routes that are safe.

What we'd like to be able to do is have a much larger scope of the information that we can collect than what is currently feasible.

So, the ways that data are being collected currently involve sending out research vessels with a number of people on them to go out to specific locations. There's some use of sensor buoys that are sent out to drift into the ocean. Some use of fixed sensors and in some cases uses of autonomous vehicles in order to go out and collect information. What we'd like to be able to do is have a much larger scope of the information that we can collect than what is currently feasible.

What's currently being collected is in a much narrower space in the entire ocean, so what we're interested in doing is coming up with ways of collecting that data more effectively using autonomous vehicles. Our objectives are to take autonomous vehicles — in this case I'll be talking about some other things we can do, fish-robot types of things, where we want to operate in dynamic spaces, so they're not very well-known, they're changing very quickly. We want to be able to send them out for a period of months so that we don't have to go out there and physically service them in order to fix parts and collect the data coming off of them, and I need to do this in a way that is energy-efficient so we can't have sensors that are gonna require a lot of power in order to run.

We won't be able to send down live video cameras because they'll need high-powered lighting in order to show what's going on. We don't have GPS; you don't have a tether for power. We want to be able to achieve these tasks again over a large space which might be in open water, or spaces that I can be close to the bottom or near objects of interest. So how are we addressing this? What we look at in my group, and with some of our colleagues, is that we take a desired engineering characteristic that we'd like to achieve, we look at where are some similar capabilities in biological systems — and similar could be a fairly broad characterization here. We then come up with hypotheses that we're going to test and we collect quantitative data for these hypotheses, then use those in mathematical models that inform our engineering design.

As it turns out, by using this approach we can also reverse the process and get better information about what's going on in biological systems, just by running backwards through our modeling process.So, what are we doing in our robots? One of the things we want to do is you want to have vehicles that are both fast and agile and some of the fastest swimmers in nature are tuna and related types of fish, and the key thing about them is they have a fairly narrow body. The first part is fairly long and fairly rigid and then there's a fairly rigid tail fin just the back. So if we take a model of a fish, or just a rigid body in water, and look at characteristics for lift and drag and mass properties, and say what would make it go fast in terms of having either a short body with the long fin or a long body with a short fin.

It turns out, if you look at this type of process relative to going fast with these characteristics, the best thing to do is to have a longish body with a short fin. So that'll give us some information about how to make our vehicles go quickly. We also want them to be maneuverable, so, again, if you look at biological systems, the most maneuverable types of fish tend to be a little bit boxier, so they have similar drag characteristics in different directions, and I have a bunch of independently actuated fans that can move fairly quickly and independently. So, what we did on our vehicles is we can just use two fins, they're independently actuated and they go completely reversed, so we can drive the vehicle backwards with them.

What this is led to in terms of design for what we've done is that we've had three primary iterations of our vehicles. The first one, which we sometimes referred to as the briefcase fish, just sort of based on its side profile there, had carbon-fiber side panels and aluminum support pieces to hold everything together, and the fins were also carbon fiber. Now, as it turned out, we didn't quite take into account the flexibility of the carbon panels and at about two feet they flexed and it sunk like a rock, which was very annoying since we should be able to predict something like that, being engineers.

Our next version, which came along shortly after that had aluminum side panels and then curved acrylic to give us lower drag, and at this point our fins on this one are 3D-printed and then coded for waterproofing. And where we're going next, which will be in the next few months, is a completely 3D-printed body that allows us to have much better control over lift and drag properties of the entire vehicle. Okay, so now we have this current vehicle, and this is a couple of them swimming in the tank in my lab, and one of them is using just the tail fin and the other one is propelling itself with the pectoral fins.

We didn't say make it move like a fish. We just said, make it move, and as it turned out it looks a lot like it's moving like a fish ...

The way that we got to the motions here is that we took our mathematical model and said, how do we make something with this mathematical model move? We didn't say make it move like a fish. We just said, make it move, and as it turned out it looks a lot like it's moving like a fish, even though I was not sure how we did it. A nice feature of this approach using this mathematical modeling is that it's based on mileage; it does capture what's going on in biology, so it does represent, if you go back and model the fish, but we also get some effects that show up that we predict will work and that do work on the robots, but they have never been observed in biology. I want to emphasize that point because we're designing engineered systems and we have engineering goals, not necessarily trying to duplicate what's going on in biology.

Now that we have a system of this type, a lot of the applications that we've mentioned, or I've mentioned in the last little bit here, involved situations where you want to operate vehicles either close to each other or close to other objects without running into them. Now that might seem like a fairly basic task, which it is, but it turns out that it is not actually the easiest thing to do. If you remember the Deepwater Horizon well-capping situation, that challenge was managed by having a remote-operated vehicle that had a live human operator running the vehicle to deal with the capping. There was a cable bringing down power and lighting for a live video feed, and they still had a really hard time doing that task without making the situation worse.

As it turns out a lot of people have developed algorithms that allow us to do what's called deconflictions, or keeping things from running into each other, but they're very constrained in terms. They either need lots of space to run or you can only do it slowly, or not in unknown environments. So if we think again about what goes on in biology and come up with some motivation from there, this is some footage from a colleague of mine who does Barn Swallow flight. These are two birds chasing each other around barn silos.This is a fairly dynamic environment and they're moving pretty quickly. Now, if you think about what's going on here, these birds are not using full vision, and they're not looking behind themselves. They're using very simple kinds of range and bearing measurements, as far as we can tell, and knowledge of speed. So we can take that idea using our process and building up this model from biological hypotheses and data, and coming up with a mathematical model, and build an idea of, given these types of measurements and an idea of the rules of the road, come up with algorithms that allow us to predict and guarantee that we won't run into things.

When we take this algorithm and go back and compare it, we have a pretty high correlation to behaviors of fish, bats and swallows.

Now, this is just a piece along the way, but it turns out so that, following the process, we can take that idea and build it into an engineered system. So what's shown here: There's two dots; the pink one corresponds to where the the pink fish on top is going and the green one is for where the green fish wants to go. If they didn't do a deconfliction, they'd run into each other in the middle, which you'll see here it's going to detect they're going to run into each other at a certain point, based on range and bearing; the same kinds of things that we think are going on in the animals.

So, they'll detect at a certain point that they're going to collide with one another, but then they redo their algorithm based on this idea of these rules of the road, and move around each other and continue as soon as they can back on their original path. You'll see that here, in a second, then they follow along their way. Now, I should point out that what's going on here, this is a very simplified environment — it is not the whole solution to the the challenge that we're trying to address in terms of not running into things and being able to do that on a very tight space or time scale. The key thing I want to emphasize here is that this is being built upon this idea, this iterative process of building these models that are based on biological hypotheses, coming up with mathematical models, and then using those to inform engineering design and also go back and verify what's potentially going on in biology.

As it turns out, when we take this algorithm and go back and compare it, we have a pretty high correlation to behaviors of fish, bats and swallows. So, what's next? Well, we're interested in taking the basic building blocks and coming up with more effective designs for systems. They are just very basic building blocks, but if we can get these pieces to work, and I think that the results are showing that we can get them to work and we can build on them in a systematic way, as opposed to an ad hoc way, we will be able to have better methods for doing environmental monitoring tasks, as well as things like hazardous waste cleanup, disaster response and planetary exploration. We're looking forward to seeing how this process evolves, and I hope that you are too. Thank you.

Kristi Morgansen


Speaker Bio

Kristi Morgansen received a BS and an MS in mechanical engineering from Boston University, respectively in 1993 and 1994, an SM in applied mathematics in 1996 from Harvard University and a PhD in Engineering Sciences in 1999 from Harvard University. Until joining the University of Washington, she was first a postdoctoral scholar and then a senior research fellow in Control and Dynamical Systems at the California Institute of Technology. She joined the Department of Aeronautics and Astronautics in the summer of 2002. Read more