This talk was conducted at the AI Accelerator Summit, and featured:
- Barbara Mazzolei, Associate Director for Robotics, Istituto Italiano di Tecnologia
- Yasmine Meroz, Principal Investigator, Tel Aviv University
- Emanuela Del Dottore, Postdoctoral Researcher, Istituto Italiano di Tecnologia
- Isabella Fiorello, Postdoctoral Researcher, Istituto Italiano di Tecnologia
- Maria Balk, Scientific Researcher, Helmholtz-Zentrum Hereon
- Ourania (Rania) Giannopoulou, Postdoctoral Researcher, GrowBot
- Lucia Nasti, Postdoctoral Researcher in Mathematics, GrowBot
Listen to this talk on our podcast Women in AI: Breaking the Mold.
The role of plants in bio-inspired robotics
There’s recently been a lot of talk about big data and AI, and that's great. But it's even more exciting if we mix it up with robotics.
That's what we primarily represent over here. We are scientists who are primarily working on a novel approach to designing robots, which is known as bio-inspired robotics.
Bio-inspired robotics refers to systems that take inspiration from nature in design principles and control principles. And then you want to deploy these systems in natural environments. We create an environment where robots are ubiquitous with us so that we can use them for assistive purposes or anything else you can imagine that doesn't exist in this day.
One of the things that has happened in bio-inspired robotics is that we’re now taking inspiration from one of the most unusual organisms: plants.
Plants represent systems that we think are sessile and don't interact with the environment or do nothing for the environment. But actually, they interact much better, which is why they’re able to colonize a bigger portion of the Earth than in fact, humans can.
So, here we have scientists who are working on a European-funded project which is known as GrowBot, where we’re trying to develop a novel, growing bio-inspired robot.
I'd like to very quickly introduce our panelists.
First of all, we have Yasmine Meroz from Tel Aviv University. She did her PhD in Chemical Physics at Tel Aviv University, and went on to do a postdoc in Condensed Matter Physics at the Weizmann Institute in Israel. She continued a postdoc at Harvard University in Applied Mathematics, focusing on memory phenomena of biological systems. Right now, she's a senior lecturer at Tel Aviv University.
Following her, we have Emanuela Del Dottore, who has a Ph.D. from Scuola Superiore Sant'Anna, which is also where I’m from. She did her Ph.D. in Biorobotics. Her research interests include plant-inspired robotics and the development and implementation of control and decision-making strategies for bio-inspired robotics.
Following her, we have Isabella Fiorello, who is a Postdoctoral Researcher at Istituto Italiano di Tecnologia. She also received her Ph.D. in Biorobotics from Scuola Superiore Sant'Anna. She's received several national and international awards, including the Eni Award for Young Researcher of the Year.
Currently, Isabella is working on microfabricated adhesives with climbing plant-inspired microhooks for reversible anchoring to textiles, skin, and plant leaf tissues.
After her, we have Maria Balk, who received her diploma degree in Chemistry from Germany, and her Ph.D. in Material Life Sciences at the University of Potsdam. She’s currently working as a Postdoctoral Researcher at the Institute of Active Polymers in Helmholtz, Germany. She's focusing on the design of shape memory hydrogels, which are sensitive to a variety of stimuli and actuator materials.
Second last, we have Rania Giannopoulou, who’s a Postdoctoral Researcher at the Gran Sasso Science Institute in Italy. Her research deals with the development of a three-dimensional model, describing the intertwining of growing shoots.
And finally, we have Lucia Nasti, who is also a Postdoctoral Researcher from the Gran Sasso Science Institute. She received her Ph.D. in Computer Science from the University of Pisa in 2020, where she worked between the research fields of bioinformatics, systems biology, and formal methods. Her scientific interests revolve around a deep understanding of complex systems to computational approaches.
What I would like you all to understand today is why we’re looking to plants. Why are plants so interesting in the development of robotics?
I'm probably the closest thing to a biologist in this set of people. I'm not originally a biologist, but I currently study how plants interact with the environment.
Looking at a plant from an engineering point of view, we have a plant that’s equipped with sensors and has no central nervous system or brain, and yet somehow it manages to get information from its sensory system and process it in a distributed manner, and then there's an output.
There's some kind of strategic plan regarding how to grow. What is the optimal way to grow? There are a lot of interesting questions here, and I'm mostly interested in understanding this black box of computation. How does computation occur in a distributed system?
Basically, we do a lot of experiments where we have a controlled stimulus and then we follow the growth pattern of different plants. And from this, we try to reverse engineer these kinds of computational principles. So it's experiments coupled with a lot of mathematical models describing this.
There are two big things about plants that I find great. There’s 80% of the biomass that's saying something. And also, if you think about the oldest living tree, which is this old pine in California, it's almost 5,000 years old. That's insane.
If you think about the number of civilizations that have passed by in this time, and the storms and fires and whatnot that have occurred, this tree is still there. It means it's doing something “right.” We have something to learn about its strategies. So that's one thing I find really interesting about plants, and we're just beginning to scratch the surface.
And the other thing is what I mentioned before, the fact that it's a distributed system and still doing interesting, complex computations. Distributed computing is something that’s of great interest and maybe we can learn something from how nature does it.
Implementing plant biology into engineering solutions
Plants have these very interesting features. What have you guys taken from biology and actually have implemented in an engineering solution, where you’ve actually developed something that’s plant related?
I'm a Material Chemist and I’m working on the design of polymer-based systems. I'm concentrating on how we can implement active movements into polymer materials.
For example, I'm working on shape memory materials that can be sensitive to the environment. For example, they can feel changes in temperature, changes in immunity, and changes in enzyme concentration, which is similar to what plants can do.
We were able to develop artificial plant leaves, which can close their shape when the temperature has decreased and open up their shape when the temperature has increased. This is a thermomonastic behavior that’s similar to rhododendron leaves, and that you can artificially create.
During my Ph.D., I worked a lot on climbing plants at a microscale level. I actually developed new materials which were inspired by climbing plants or reversible anchoring to several structures. I took inspiration from the biomechanics and morphology of selected climbing plants.
To give you an example, I took inspiration from a plant called gallium aparine. This plant can actually anchor its body to surfaces in nature. And what I did during my research was take inspiration from the microstructure that’s present on the leaf of this plant, and mimic the morphology and the biomechanics of these microstructures.
Then I designed and fabricated this structure at a microscale level using a micro-manufacturing technique. You can actually use these devices in different fields. For example, we applied this in the climbing robots applications, but also, you can attach them to the leaf for leaf sensing and drug delivery.
There are many features from which we take inspiration. I mainly took inspiration from the anchoring, but also from the morphology and biomechanics.
Systems that are attached to actual biological organisms. In a way, we’re moving towards a new novel interaction between robots and humans.
Another thing that’s very interesting is that Emanuela, one of the first growing robots, and the growing part of the system, is being developed by you. What exactly is the mechanism that you're working on? What are the computational principles that you're using? And how are you developing control algorithms for these robots?
Plants can offer many different ideas, both in the materials, because they’ll respond to many different types of stimuli from the tissues, and also from an internal, physiological computation where they can improve their performance in terms of growth and direction.
We took inspiration from this basic but fundamental principle in plants that they move and explore different environments through their growth. This is a very different concept of movement with respect to animals, which shift all of their body from two points in space, while they’re adding new materials at extremities.
Plants have affixes on the roots or any parts where there are cell divisions and variation. It’s only this time that they move forward in the environment.
We took inspiration from this basic but effective principle to develop a growing system, recalling to the principle of the additive manufacturing processes.
We basically miniaturize with an FDM 3D printer in a four or five-centimeter scale of diameter. And what the system does is take the thermal plastic material in the commercially available 3D printer and fuse it at the tip and deposit this filament in order to increase the body layer by layer.
The body is not actually redefined. We’re basically building the body of the system in real time, so the system embeds a control that’s inspired by tropism, which is the main mechanism for plants to navigate the environment. So they grow towards the light. The shoots are attracted to the light, or they escape from negative obstacles or metals or other stuff that can kill them.
That's so very interesting. And Emanuela's raised a very interesting and important point, which is embodied intelligence, which is something that has been overlooked so far.
We consider artificial intelligence as mostly the algorithmic part. But in fact, what we’re trying to explain is that the body itself plays a big role in developing artificial intelligence, especially when we’re talking about these artificial systems and robotics. We can no longer consider robotic systems independent of algorithms. They have to work hand in hand.
Maybe Lucia or Rania can highlight what they’re working on, which is modeling and trying to understand these complex interactions, that something as simple as plants are able to do with very simple bodies, such as intertwining, and these computational capabilities.
Hello, I'm Rania, and what I've been occupied by in the GrowBot project has been finding a mathematical model for intertwining. This is a strategy where plants have evolved in order to bridge gaps among structures, climb on other structures, or attach to other plants.
Essentially, what they do is twine different stems together, and this has been shown to improve their mechanical and structural properties.
The point is that we’d like to have a mathematical model that we can understand deeply, and then translate this model into something that’s amenable for robotics use.
The main quantity that we measure is the stiffness of the material, and when we compare one shoot growing by itself or two shoots growing together, we see a very large difference. So this is an example of how we can incorporate these mathematical models and the knowledge into AI and bio-inspired robotics.
If I can add something, I think that it’s important to state that AI is playing a key role in developing predictive models in the biological system. But a limitation can be that in many cases, we don't have all the data, so we have to deal with incomplete or missing data. Therefore, we can’t rely on this kind of approach alone.
Maybe what we can do instead is combine mechanistic models or mathematical models when they are available, with the potentiality of machine learning or reinforcement learning.
I’m currently working on the application of reinforcement learning to understand the plant adaptation strategy. When we work with reinforcement learning, we need very little data so we can also generate our data in the training step. We generate data through the interaction between the agent that is the model in our case, and the environment. So perhaps this is a great opportunity for other applications.
Unlocking the potential of plant-inspired robotics and AI
When we’re talking about this completely new concept of people who are dealing with algorithms, what would you like people to take away from plant-inspired robotics, the role of AI in plant-inspired robotics, and the future of plant-inspired robotics?
For me, the most important thing is the control strategy. For example, with reinforcement learning, we’ve already observed the environment, and then we have to perform many actions to complete the task in an optimal manner.
This is what the plants do in nature, so maybe this is the key point of these two approaches together.
For me, I would say that it’s important to know that the appliance shows a diversity of different structures which are able to respond to the environment. They can respond to a diversity of different signals, and they are much more complex than what we thought before starting this project.
For me, one very important thing is the capability of plants to adapt to several structured environments. And I think we must take inspiration from there, especially from the material properties of plants.
We can mimic the morphology and biomechanics of these materials to create new materials for robotics. Actually, I can see a connection between plant-inspired robotics and artificial intelligence, especially if we think about applications for agriculture or forestry, for example.
During my research, we developed smart materials that are inspired by plants, but they can also be applied to plants. Maybe in the future, these materials can work together with artificial intelligence that can create a network to understand when to give water or pesticides to plants. We can envision these kinds of devices in a vineyard or in a forest.
I would agree with Isabella. We can use all this knowledge to actually create new materials. We can use them for robots or environmental monitoring where you need something that can survive in a particular environment.
We’ve already seen all these new robots that are for self-manipulators, so I think that there are many options where we can apply this knowledge.
Plants are an excellent model of embodied intelligence, and an excellent example of a distributed system. We have computation distributed all along their bodies so that we can achieve very high complexity that is displayed in plants and in their phenotypic expression. But perhaps it’s reached with simple rules that are implemented locally.
So, perhaps this isn’t a very new concept, but this model can renew the paradigm of modularity, distributed intelligence, and embodiment.
I'd focus on one thing, which is just to keep in mind how fundamentally different plants are from animal systems or classic robotic systems.
We all know those very cool movies from Boston Dynamics where you have a robot that pushes Atlas or Spot over and it manages to somehow stay put. So how does posture control work?
I know what my body consists of, and if my senses feel forces that they're not supposed to feel, we can interpret this as an external force. And then we push back. That's also how we stand when someone pushes us as human beings.
But what happens with a plant or a growing robot? The morphology changes continuously. This means that the whole fundamental concept that we know about in posture control, which is very well understood and very easy in classic robotics, basically means we can't use the existing control system.
We have to chuck everything out of the window and start from scratch, and since it's hard to find solutions for these things, we need to learn through plants. How do plants find solutions for this? So if posture control is already difficult, next time you see a tree and you see that it's straight, you can marvel at that. But this also means that everything else is complicated and conceptionally very different.
In animals, development and behavior/movement are two separate things. But in plants, we can’t take these two things apart. They're always together. Through development, which is growth, that's your way of manifesting your behavior or moving. And this means that everything is different. All the computational processes need to be different.
It's a goldmine of open questions, and that's what makes it fun for everyone here involved.