Stanford Robotics Seminar ENGR319 | Winter 2025 | Embodied Intelligence

0:10 It's my honor to come over here to talk a little bit about embodied intelligence. it's something my lab has been thinking of over the past few like monthsish. we we generally work with morphing matter. We recently started to think about you know the possibility of leveraging morphing materials and mechanism basically to think about embodied intelligence from from our perspective. So it embodied intelligence I'm sure you guys have heard of it under different context perhaps it means many different things is vaguely defined at this moment. So from from the context of this talk we are pretty much talking about some level of programmability or decision making through hardware system especially in our case through shape changing and tunable materials and structures. So people argue those systems are are useful from different perspectives. You could you could argue this kind of purely mechanical systems that can do computation or some sort of actuation I is good for cyber physical cyber security. So the idea is there are certain important lock you can only unlock when you are physically being present. people also articulate those mechanium computers and actuators are useful for for robotics for especially electronic free robotics for extreme conditions. and and unique environment. and we in our lab also try to think about how such mechanical systems can be basically leveraged in sustainable application.

2:01 For example, robots working in the field be able to harvest the ambient energy like the sunlight, the wind, the moisture fluctuation from the environment and automatically conduct some robotic tasks and maybe eventually degrade into the field. So with with this in mind, I kind of stitched together a presentation to talk through again some of our team's recent recent reflection in terms of what are the unique sort of applications or unique technical perspectives when when we discuss embodied intelligence again through the lens of morphing materials and structure. So there are four aspects I want to b like touch upon with a bunch of examples that that u that are that are from our lab. start with the first one. So embodied intelligence in the context of physics plus algorithm. We design a lot of morphing materials. So when we think about those we really think about it as material system that carry hidden forces that are usually calculated or predicted based on some computational algorithm. That's why we think morphing materials are physics plus algorithm. So usually you deal with the certain physical morphing mechanism and you think through how do you compute them, model them and how do you do inverse design by leveraging these kind of a physical phenomenon.

3:29 So through the years we've been looking into different material systems that could leverage this mindset for design and engineering all the way from ship memory plastic to pneumatic elasic system to even you know clay and even edible gels and most of the project share the commonality of physics plus some sort of a computational design behind it. So here are some quick examples. So for example, we once looked into basically the idea of u residual stress release inside thermoplastic can be seen as a way to engineer self folding and self assembled structure. So essentially we're using basically FDM printers fuse deposition printing method to extrude the ship memory polymer and lock some residual stress in place while you are printing them and then later if you try to basically heat it up the stress will be released and the filament tends to shrink. So if you print this shrinkable shrinkable material on top of a non-shrinkable material, you can create a effective blayer that will bend when you introduce extra heat to to it after the printing. So this FM simulation is basically trying to to simulate what what's what's happening.

4:55 and we then knowing a bit of the physics then we we know you know we can do some sort of a bending actuator. We we move on to the computational side. Try to leverage a origami algorithm to compute the local bending angle needed accumulated bending angle needed at each hinge area so that we can generate a G-code with prescribed residual stress at each local folding area. And we print everything flat, but once you put it in water, heat it up, it can still fold back into the Stanford bunny shape.

5:31 And with that you can start to design all interesting shapes that can self assemble from a flat piece. a flat piece when placed on a hot water surface turn into a boat or a disc turn into a chair. So this is little bit inspired by IKEA flat pack furniture to save packaging and shipping space. But once it's on site it could you know you could assemble them but in this case it could assemble itself. and if you're interested in the computational part of it, so this is a very simple and standard basically origami flattening algorithm. You simplify a 3D model into finite numbers of flat faces and you calculate the ang folding angle needed at each hinge. but here we need to do something different from a traditional origami model in a sense. We need to compensate the the width needed because it's a smart material. any material for a bending to happen need a certain width versus the traditional origami model because based on very thin paper that width is ignored. So we had to do basically some offset in in the algorithm side to u take into account the the width of each bending region and the width is actually indeed different depends on what is the bending angle needed anyway. So this eventually gave us a software platform where you can basically input the 3D model, simplify it into a into like I said flat faces and calculate the bending angle, generate the G-code from it, print a flat piece and then this is showing once you put it in hot water. So the flat piece can self fold into the spiny shape. so this is based on this very simple blayer blayer phenomenon.

7:20 and and fundamentally this is as I mentioned is related to right the residual stress release of a thermoplastic and the residual stress was built in during the printing process. So instead of doing blayer we can explore a different physical mechanism which is basically leveraging the local linear shrinkage rate. so in this case is not a blayer anymore. Basically we print a single layer but depends on the print speed or the the layer thickness we could embed different residual stress amount at different local area. You can print a disc once you heat it up it morph into a dome for example or a cone and you can you can do computation now to make more arbitrary shape. If you're a little bit into computer graphics, you might recognize now we can basically analyze shape that are non-developable. So origami shape are bit easier to handle geometrically speaking because they're developable. So here we are basically developed a algorithm that deal with non-developable shapes such like a human face or or a very specific mountain terrain condition. Basically, you input a 3D model, it flattens it and generate all the G-code and you see the G-code here. U the color indicates the the the stress or the strain level. and we can also simulate how it would morph into 3D. And again, you would print a flat piece in this case a flat disc u that morph into a face after you give it extra heat.

8:59 so we had a lot of fun with thermoplastic and then we moved on to look look at some other physical phenomenons as well. So this is a representative one that I love a lot. This one look at a physical phenomenon called differential swelling that tuned by by the surface pattern. So so imagine this is a swellable gel. you put it in water it just uniformly grow thicker and wider. but but now if we introduce basically a sub med sub millimeter scale grooves on one side of the gel it'll it'll impact the swelling rate and ratio across the thickness. So the side with the groove will ends up swell more slowly than the side without the groove. So if now you basically let it swell in a solvent in this case right a starch gel in water you will see a flat piece bends up.

9:57 Essentially what you get is again it's still a bending actuator like what I showed with the thermoplastic but the the physics going on is very different. So this is a differential swelling and the previous one is a residual stress release. So once we get to know a little bit of the physical phenomenon then again we plug it into some computational pipeline. So in this case we had to develop a pipeline basically to to tell us how the growth are are are supposed to be distributed right. So you can see if our end goal is to get a rose flower, we basically produce a flat stamp in this case with groove pattern on it and then we stamp on a on a swellable gel to create the surface texture on the gel and that's what what we are showing at the bottom. So so basically a flat piece if you put it in a solvent in this case it automatically start to swell and wrap it wrap itself up into a rose flower.

10:58 and we also so this phenomena is really straightforward and no magic chemistry you're right as a purely structure modulated morph mechanism but it's also on the other hand powerful because it's it's applicable broadly applicable to soft matter that could swell in a solvent per se. It could be for example your dumpling to some extent your dump your dumpling cover being boiled in pure water. It also could be some sort of elastor or silicone sheets being swollen in a in a polar organic solvent. So we showed in a paper that we can even use this to make fun morphing food. So basically this case you can you can make a grooved flat pasta when you they're all flat pack you IKEA idea save a lot of packaging space for food. but once you started to cook them they take on different shapes that they're supposed to be paired with different sauce with different texture. But obviously this could go beyond you know food. If you're into biomedical for example you could make basically gel-based robots with this. So I'd like to go a little bit deeper into into this mechanism. Again I said the groove is a really unique feature that tune the tune the morphing behavior. So initially you get a flat piece into water. By the way the left is a experiment but right is simulation in this video. So you you you can see it started to bend towards the groove side and then it gradually reach equilibrium and actually now if you take it and go flat. Now if you take it out of the water it bends the opposite side. It's a very dynamic phenomenon going on. and interestingly this you know if for a moment if you like ignore the physics this is geometrically quite relevant to the pattern of the groove. So it turns out the the width of the groove the gap between grooves and also the thickness of the of the groove will will impact how much it bend.

13:11 So there's a relationship you can derive experimentally and then you can do regression right to get a sense how basically the curvature will be tunable based on the geometry of the groove. So once we get that we went into our kind of comfortable zone of doing computational design with it. So we based on the experiment calculated we we concluded one group contribute roughly to 12.4 degree of the bending angle that basically means you get two two groups you get like 24.8 degree and and you can imagine I added one group I contribute to like 20 20 some degree.

13:55 and then I add another one, I contribute to 48. I added a third one, I contributed to more. So that's how you could potentially design a flat sheets that can curve into any arbitrary shape. but but that's tedious. So there in graphic world, computational graphic world, I do have a laser actually. Yeah, in computational graphic world, you can do this kind of more holistically. So you could basically given a target shape you want your flat sheet to morph into you could calculate the scalar field and then automatically derive the locations where you want to introduce the goof. So so from that you could go back to the the complicated roseflower case I just mentioned. So if you know your target 3D shape like a rose flower, you could unroll it to get the flat disc, you could then in parallel also derive the scalar field. Then then with that you can map the grooves to the specific location of this flat disc. Then we basically generate a 3D printed mode which function as the stamp. We use that to either mode cast or stamp on a flat sheet to get a a morphing disc.

15:11 Basically that will eventually morph into the rose flower if you put it into a swollen a swollen medium. so with that you can go more complicated with the with the with the morphology right. So I already showed the flower. So the one at the bottom is an interesting gripper. So it was flat at the beginning. You you dip it in solvent. You can grab something because it will bend once you take it out. like I said it's dynamic process. it will reach its own equilibrium and try to recover back to flat. That's a moment you can release whatever you are grabbing. Yeah. So basically you can make a semi-autonomous gripper with with with just this groove feature and this can be applicable for a lot of gels. So in this case we're using PDMS. If you are familiar with micrfluidics and some biomedical soft robots you know PDMS is really basically compatible with the human body.

16:10 So the yeah so this is this is the fun application we did. So these are dough. So actually semolina flour based based the dough and we are doing the stamping on top of the dough and once you boil them they can take on different 3D shapes. So this is the morphing pasta example on the left. So these are showing some fin element simulation and then and then you also seeing some real sample of the pasta and yeah we started to you know if you ask why why pasta so it's it's really fun and it's a very universal phenomenon right flower based doughish food can swell in in your kitchen as you boil them but yeah we we if you also do a simple calculation macaroni pasta they're like about 67% of the space is used to pack air. So if you make pasta that could sell basically be packed flat, you can save a lot of packaging space.

17:09 So we actually brought this to Italy and we know people in Italy basically 1% of the green gas emission is generated by by consumption of pasta. So we eventually convinced that yeah Barilla who you know the biggest pasta manufacturer maybe in the world definitely in in Europe to collaborate on the project. So yeah it's pretty fun. okay so so that's some sort of intelligence related to how physics combined with algorithm. I want to share my second point that is that in body intelligence can be sort of amplified by by combining materials and mechanism. So I want to believe what I just showed are mostly materials. they are passive, right?

18:01 They're gels, they are thermoplastic. They can do certain things, but they're also quite limited. and some PhDs in in my lab through different kinds of project started to bring up the idea that if you can combine smart material with with some structural intelligence or mechanisms per se, you could you could have more programmability and controllability in in morphing in morphing system. So when I say smart material, you know, yeah, temperature stimuli responsive, it'll bend, it will shrink, it'll change stiffness, that's about it. But there's no like degree freedoms. There's there there's no there's no sophisticated kind of structural programmability into it. so what I just played on this video is a stiffness changing material. So this is again a shape memory polymer but if you introduce in this case we we had a almost invisible heating wire embedded in but basically if you heat it up it becomes really soft otherwise stiff and you can it remember it shape so automatically recover back to straight if there's no external constraint. So, so, so, so my students started to basically try to combine this smart material into complying mechanism to do something that's fancier or something the material itself cannot do.

19:27 So, the whole project is about how you could basically make reprogrammable compliant meta structure that has reconfigur reconfigurability around all six degree freedoms. sounds complicated but yeah basically it's a it's a compliant mechanism but you can selectively think about it soften or harden certain rod within the within the compliant mechanism. The ones that get softened basically free that degree freedom. So you can rationally design it to achieve the degree freedoms and reconfigurability you want. and this is the basic design principle.

20:06 Again we know how to address the flexure. You can do basically screw algebra to figure out the degree freedom depends on how you align those rod rod when it's stiff is a constraint right and then we we we can define which rod is that so-called smart rod that could reconfigure its own stiffness and then we look at the degree freedom of the entire system as the as the output. So the idea is when it's stiff right you get one degree freedom status but when it's soft and then suddenly it can bend you increase the degree freedom and this is one example we showed in that paper this is basically this is completely stiff both rotation locked and this is one is up down rotation unlocked so you could yeah the palm can basically right push inside outside and this is this left right rotation unlocked. So now you can do left and right but not up and down. and then if all of them unlocked and now you have full degree freedom. we talk about using this for for example potentially for rehabilitation or some haptic device. and one thing that we ourselves are proud of is we introduce a rational design pipeline basically like algorithm that allow us to design systems that have all six degree freedom fully reconfigurable. You could you could ask for what you want.

21:39 For example, only X translation is flexible all other locked. This is depends on how you again arrange the stiffens changing rod versus the other non-stiffens changing rod. the chart is basically showing when you heat it up versus cool it down you you get a drastic force force difference for the same structure. So again the first column shows a device that only has X translation re reconfigurable in terms of its degree freedom and the second column is Y translation programmable etc etc for all the sixderee freedom and this is this is a video showing a bit of the same concept.

22:25 So again, X translation now is freed because we heated up the selective wires our algorithm told us to to heat up. and then this is a Y translation being being freed. It's a it's already a different mode. this is Z translation being freed. This device really doesn't have any functional meaning. We just wanted to show the generalizability of the of the basically the the design method and this is X rotation. So I'm going to skip the video but I think you get the idea. so so yeah so once we understand how to design the deg freedom as as we want we can look at some more specific applications. So again I've already mentioned we picture these type of exoskeleton devices can be used for example to to constrain certain degree freedom to do like a haptic virtual h kinetic feedback for example if I try to turn a door knob door knob I can just free that rotationary degree freedom while lock the other ones.

23:41 and we also picture we can basically right not fully heat up the wire but heat heat up the wire partially with the PWM so that we can use it to to do tunable rehabilitation training. yeah and there are perhaps many other applications designers could think of. but but I really see this project as a way to think about the combination of materials and mechanism and it's very mechanical engineer friendly because yeah my student if they come from engineering background they love mechanisms. and these are these are some ongoing work. So we are trying to explore some more complex sort of compliant mechanisms that could even have some computation or logical logical logical behaviors built in. So this is just a door ledge but it echoes back to my idea of physical cyber security.

24:39 So for this door latch is purely mechanical. You could but but you need to introduce three sequential motions to open it or to unlock it. So in this case you have to do a yeah like Z translation first and follow with a rotation follow with a a Y translation in order to basically unlock it. And this this again is designed by by this inverse pipeline we introduced in the paper. This is just video to show maybe the left one is a little bit more clear. So you can see the hand has to do that three sequential motion exclusively in order to basically open the door door lock. We were very excited because this is really getting into that logical computation you could do with a compliant mechanism. And if we embed the smart rod right we can reprogram the possible logic you need. you know this time you need to do this sequence to open the door but next time if you soften a rod you have to do a different sequence. So that's that getting into that magical cyber physical security part of it. We were also thinking again this is semiconceptual the design is not even mature but the idea of leveraging this type of reprogrammable compliant mechanism to create multimodel robot. So this one you think about the blue one as one groups of flexure that can be softened versus the purple one as the second group and depends on which group you soften or harden. You could reconfigure this robot either be a walker or be a swimmer.

26:19 And this is just a simulation. We actually have not been able to implement this with the physical prototype. This is the walking mode and this is the this is the swimming mode. No idea whether they're gonna work. at least the compliant mechanism part is correct. we are pretty sure in terms of the the the swimming or walking we we we have a lot of optimization to do. So this is a different example. Again, I'm just stitching project together to make the point that material in combination with mechanism could get to could get us a little bit further than pure material-driven system. So this case we are having in this case the smart material component is the shape memory alloy in the form of this coil embedded in this in this rubber structure that's that's rendered in white. and the mechanism part is this is this white frame and also this blue part. So it's a basically like a bstable mechanism.

27:27 So you you can see the fabrication process. Basically we have a 3D printed frame. Then we stretch a very thin membrane and then we basically fix the stretched membrane into into the frame because of this elastic energy we introduced into the frame it is bstable. Basically if you try to stretch it it can quickly flip to the other side. but this is purely passive. Then by embedding this ship memory alloy, we could basically right leveraging the shape memory alloy controlled by electricity to quickly flip the side of this this bending actuator per se.

28:08 and this is yeah this is what what is what it's doing. basically again you pump in you give it a current signal it it quickly flip. If you have played with ship memory alloy, you know it's a little bit notoriously slow and pretty weak. but but by combining that into a bstable mechanism, you can actually turn a very slow actuator into a pretty fast and hopefully more practical one for building robots. So and we are also showing this this actuator has pretty good repeatability. In the lab we tried to run it for more than 500 times.

28:48 basically circle back and forth. It's still behaving pretty well. and these are some of the robots we are able to we were able to build. So this again so this is the amphibian robot. Basically the body is made of two of those bstable actuator. It can you this is the walking mode and then and this is at the bottom you flip the two actuator at the body you make it into a swimmer. And this very tiny body is showing this is walking.

29:19 This is getting into the water. And once it's in water, it flip its body and the legs become the wing. Is the wing? Yeah, the wing. and started to basically pedal. And because I mentioned it's very fast due to the bicycle mechanism, you can make a jumping robot with ship memory alloy. I thought is pretty magic because again the actuator itself is very weak and slow. So you you could also basically reconfigure the body shape of the of the of the robot. So in this case you could either basically you can turn a walker into a roller if it has to roll down the hill. They they form a ball shape. Otherwise they can they can just bend up and down to walk itself.

30:09 yeah that's that continue with my third point. So we also I also think you know we discussed this with many colleagues how far we can push with basically mechanical system. I I visited the robotic center and many robotic lab this morning at at Stanford as well. I definitely saw a lot of hard robot basically robotic arm driven by by motors per se with a very classic robust design and ultimately those machines give us the the most pre most precise control right the amount of torque that's more ideal for heavy duty heavy duty things and it's fully controllable in a sense you can easily plug in machine learning or you know like sophisticated control policy to control those robot and and mechanical systems or material-driven systems are the opposite. They're slow, they're not precise. they're very weak. absolutely in the field of soft robots and basically mechanical intelligence. We discuss a lot whether they're a killer app at all for for mechanical systems. I think we more or less started to reach the conclusion maybe there's something that can combine both mechanical intelligence and computational intelligence and reach somewhere maybe neither could achieve by itself. This is super experimental exploratory thoughts. I actually can't really think about very convincing robotic system that shows both very perfectly. But yeah, but again here when I see computational intelligence, I mean those fully programmable robot system and mechanical intelligence is closer to what I just showed. So so the question is can we combine combine these two not sure but I I'm showing some earlier experimentation maybe give some hints at least from our our perspective. So this is example talking about the design of mesh robot. maybe I've seen some mesh robots around is by different name variable geometry trust trust robot right or or or like looped looped system but ultimately yeah so they made a bunch of trust and they are connected by node. So if we try to make a morphologically really complex mesh robot like this in this case we are thinking of basically pneumatic actuator pneumatic linear actuator you you you you have to have lots of separate control units you have to have lot of actuator lot of wiring. So if you want to independently control the basically each beam or each truss per se. So for a lobster robot like this you you need a you need a 67 control modules. So we yeah we didn't perhaps want to do that. So we got quite inspired by actually this is during my visit to Northwestern Robotic Institute.

33:16 There are some biomechanics people told me basically you know there's like this muscle synergy concept multiple muscle groups are supposed to be grouped and kind of work together to achieve task more efficiently. So we started to look at this idea and you know think maybe there are ways we could also mechanically group the the master actuator together. So instead of 67 control module maybe we could just have like a much fewer numbers of module in in this example just three control module but allow the mesh robot to to do equally sophisticated things. So we ended up yeah doing that. So basically you can imagine if we if we need to group them each node may have to carry multiple air channels. So we developed a optimization algorithm basically to automatically generate the the the air channel connection to and the the the the basic idea is is like this is very simple right you have three groups for example represent by different color and depends on which group you you deflate or inflate you could get quite different actuation behavior. So the design problem indeed is a little bit complex. You have to think about which beam to be grouped with which one. You also have to think about the control sequence right of of all those beams. basically that control policy aspect. and yeah this this is one of the example again that lobster designed by the the the idea I just mentioned and we also tried to implement the physical version and you can see this lobster is only carrying two tubes in this case. So basically this is grouped into I believe two groups.

35:12 and I mentioned the design task is kind of complex if you try to do it manually. You have to think about the grouping. also each actuator control how how much contraction ratio you want out of each actuator. Also the the control sequence. This is the control policy. Which group will be triggered at the same time? Things like that. Now if you want your robot to do multiple kind of tasks, this is getting even more complicated. What if I want it to turn around while lowering its body? So it becomes kind of multi objective design design task. Eventually we decided a simulator and manual design. Manual design means manually select beams in a software system is is not really practical. So we started to get into basically optimization method. This was a earlier version basically using genetic algorithm to figure out how to do multi multi-objective optimization.

36:13 Basically given all the input design factor we can figure out how to group the channels and how to control the the inflation deflation. and later we also explored how to basically use reinforcement learning to handle similar tasks. and recently we started to also kind of leverage generative algorithm to come up with the robot design as well. and yeah this is this is what we had out of the genetic algorithm. Again the idea of this like four-legg robot being able to do multiple tasks through the optimization method I I showed in the previous page.

36:52 So it depends on basically the the channel grouping and also the control signal. You could you could let the robot do different things while not changing the grouping obviously. so this is showing yeah walking, rotating, tilting, lowering the body and things like that. and you can you can why why we are interested in such complicated mesh anyway? I saw many mesh robot much simpler right maybe like a single unit.

37:23 I know colleagues over at the mechanical engineering department at Stanford they they had some really cool robot made of one or two unit basically maybe like a 10 truss but we were interested in hundreds of trusses partially because we can do it with a with an algorithm but also partially because we were interested in understanding how the morphological complexity could bring some unique design spaces. For example, if you think about your entire helmet is a robot. Can you possibly morph morph the shape of the helmet to afford different tasks?

38:00 Maybe depends on the scenario when you wear it or different people with different head shape wear the same same helmet. Can you morph it to fit the different needs? So this itself is a morphing, right? is a again a multiobjective basically task you can frame it. and this is also at the bottom we are just showing we could actually build it. So this this is the simulation result. When we actually build one we were tracking basically how precise each node could u basically displace and whether that match our simulation or not. but but yeah from a design perspective you could really start to think about all complex robots or robotic arms manipulators even daily products you can build with it. So we we we are speculating some more complex things but but although buildable hardware wise not sure. so this was a proposal we wrote for a hospital. Basically like a morphing a morphing mesh bed that automatically fit the body shape or automatically adjusts the patient's body posture because they get tired being fixed in one one one pose. you could also imagine if the mesh gets complex enough you could really physically render different physical object in a mixed reality mixed reality environment or in this case yeah basically give people different experiences. Say you were playing a game, you you turn into a bird. How how that physical mesh can morph not only shape is morphing, right?

39:47 Your center of gravity will be shifted as you are transforming the physical piece you are carrying as well. okay so my last point I want to articulate yeah why why embodied intelligence is is is interesting what's good for so I started with that question I I guess I didn't give an answer but these are u definitely some more tinkering side of it my my lab was also very interested in looking at basically how purely ambient energy powered system could play a role in the field. Imagine field robot deployed massively in a distributed fashion and eventually degrade into the into the field. and this could be convincing why you do not want electronics to be part of it.

40:40 Why you want really cheap things to be made? Why you maybe want minimum intelligence but be able to basically execute the task as it's intended to. So so yeah so we started to play with this concept called ecological physical AI or ecological morphing matter. So the idea again so they are some sort of a mechanically intelligent robot but they are powered by the environment. So the environment give us the energy whether this whether it's sun or again moisture fluctuation or even geothermal energy. and you design a physical system that could smartly harvest those energy and also maybe some sort of a mechanical computational unit that could take this input signal, process it, make decisions in terms of what you use those energy to do on the actuation side. So that's a framework we started to develop. and this this is not published yet, but it's a it represents some of our thoughts behind. So how you perceive the basically natural stimuli as some sort of signal and how do you build mechanical devices that rationally filter and process those signal and how you couple that with morphing materials and structure to do useful things. So for example like a facade piece like a window basically mounted on the window could responds to very specific moisture and and like a light condition. automatic open and close to tune the to to tune the light permission on your window or how a system built to basically har energy and do robotic work in the field to manage a garden. this is a lot to digest. I didn't mean to basically let the audience fully digest it but but just visually give you an impression where we are going. So you could you could use the lens of import embodied intelligence to build artificial systems but also to interpret how some of the natural systems work.

42:47 So yeah I'm going to skip this not getting into details. So so there here is example in terms of again purely passive little bit smart not quite smart things but play a role in in natural condition. So this project is inspired by a seed from nature. It's called eodium. Basically it has a coiled body that's responsive to moisture. So in this case basically ring comes the coil will unwind and that unwinding generate a rotationary motion and as a result generate a thrust force downwards to push the tip into into the soil. I'll play again. Again the coil unwind the whole thing rotate and then push the tip into the ground and tip is where the seed indeed is embedded in. So this is also a little bit like your power screw.

43:41 and we basically got inspired and started to try to make biomimeatic version of that. So on the right this is this is u made in the lab. it's actually the raw material is just a wood veneer, white oak or maple wood veneer. But you can now if you can engineer your own version, you can carry the target seeds you want. So we were very interested in reforestation. So actually right now we're working with the Cal Calire Department and also over the east coast with the Cornell forest try to try to actually engineer seed carriers for specific tree species. So the idea is that you could you could carry many many of those anyway. They're biodegradable. They are made of cheap wood veneers and drop them onto the ground. Carry the target payload and once ring comes the self drill into the ground. Drilling into the ground is very helpful to improve the the germination of the seeds you are interested in.

44:42 and this is this is a video showing showing basically like a earlier field test we did at Pittsburgh in this one and again it's really really early stage test as you can see with a really small box only carry tens of seeds but we were able to show we can drop them onto the ground and response to the rain in the afternoon and self drill into the ground. So through the project we also got to collaborate with lot of ecologists and we learned basically reforestation again is still a very big challenge because right now people basically have to do seedlings and then they dig holes they manually go up to the mountains put the seedlings in. It's a lot of resources and it's also very very dangerous if you have to go to like a sharp slopes to to do seeding. and trees are harder much harder to germinate than vegetables. So it makes a lot more sense for us to apply this technology to improve germination of tree seeds.

45:44 And for the engineering side, we're basically looking into how we could possibly do things better than nature. I'm not sure if you guys noticed the one in natural system has a single tail. So later we started to add more tails, play with the kind of improvement of the the torque of the coiled actuator. So there are a lot of basically engineer optimization work going on in the lab right now and this is part of the result of the field test. Basically we're we're we're showing we got some initial success for outdoor test.

46:21 and we also thought about how we could engineer those you know not so smart carrier to be a little bit smarter in this case by potentially adding sensing component to it and the wireless communication as well. and you can even possibly network them to to to do things at a little bit of a network scale. this is the earlier demo to show you could basically equip a antenna to do a wireless communication. The vision is you will be able to monitor some conditions maybe under underground condition and then send those information back to a drone that flies by later on.

47:09 so yeah so I have oh this slides yeah we we talk about you know this could be really helpful for reforestation we are also discussing with farmers for fertilizer basically precision agriculture to see fertilizer uses and one interesting part if you're into space so a lot of the earlier study of seed of this kind were sponsored by space agencies because they were also interested in basically environmental friendly completely like energy free to some extent electronic free system that can be used to do soil sampling in outer space because the system although the one we developed is harvesting the moisture fluctuation but the same mechanical principle can be used for example to harvest thermal fluctuation but do the drilling only if you change the moisture sensitive actuator into a thermal sensitive material. So, so yeah, this there could be some really cool applications for outer space exploration. that maybe is also a killer app of mechanical computer.

48:23 so this last example very short again follow the same threads of thoughts how mechanical computer can be a good use in outdoor. So another team in my lab is basically envisioning a fully automated garden from seeding to watering to like fertilizing for for like extreme weather condition protection everything done without electricity and purely with mechanical mechanical systems. and we see them as robots because they have yeah they have sensing they have energy harvesting they also have actuation. The the idea is that you basically build multiple stimuli responsive pump and and valve components all mechanical. The pump is a way to basically harvest ambient energy and turn them into pneumatic energy basically compressed air. so for example this video shows how this component is designed to harvest the thermal fluctuation in the air and accumulate compressed air into a chamber. So basically it's a it's a thermally responsive pump right and the bottom shows a moisture valve. What it means is that when the moisture comes the pneumatic channel will open up.

49:51 That's why it's a moisture sensitive valve. It's done by a very classic mechanical mechanism called kink valve. the idea is that when moisture comes, there's some moisture sensitive hydrogel get swollen and and then basically close the kink valve by literally kink the tube. so so so that's just two examples and eventually we are able to construct a library of stimuli responsive energy harvesting devices and valves. So you could harvest thermal fluctuation moisture fluctuation. You can also harvest kinetic energy such as wind and hydro pump pipeline them into a compressed air. And you can also design valves that responds to different environmental stimuli as well. So with all those in place we started to construct pneumatic circuits that take in account of the environmental condition. In this case this is getting very complex if you're not I guess super into pneumatic actuators. This maybe even a little bit trickier to read but but example make life easier. So yeah. So what I just showed is basically like a dispenser that automatically get triggered when the weather and the moisture condition in the environment is right and then and then after a certain amount of time a valve will open up to do watering and then after a certain amount of time predefined again fertilizer comes in and then when you detect the right wind condition and also it's cold enough a wind blocker will come out to protect the plants.

51:32 I hope you guys are paying attention to how I'm describing it. There are a lot of if and when and and end condition. So basically there are a lot of like logic being programmed into the into the hardware system when we think about this kind of like fully self-regulated plant plant device per se. Okay. So to wrap up my talk we discussed over lunch hey maybe the future super powerful devices has the kind of hybrid intelligence both from the machine and also from the material intelligence or you can think about it as a as a system that that's hybrid right you have the chips you have the artificial mechanisms and also you could have biological mechanisms MS or biological materials living or dead being part of it as well. That's at least that's the space the intellectual space my lab is trying to tinkering around at the moment. And yeah, thanks for for your attention.

52:46 >> Any questions from the audience? >> Yes. So for the materials you said you were measuring basically the width and the group and properties like that. Are you considering like the the surface area of diffusion to whatever solid you will whatever solution you're putting it into >> your time as a Martin pasta? >> Yeah. That for example. >> Yeah. Yeah. So we when we do final element analysis on those structure we basically plug in all those factors. But in a sense not like analytical solution we do numerical simulation right? would measure the diffusion ratio, the change in modulus and just just let the numerical simulation does its magic per se. So it'll automatic right take account basically it actually renders how the water diffuse into it with or without the groove in that case. Yes, the surface contact area definitely play a really important role. but when we do that very simplified like group geometry matching to to the contributed bending angle per per group that we we did not but we so but the so I I only have half of the story the whole story is the group we know play a very dominant role but not the absolute role.

54:09 So there are a lot of nonlinearity to the morphing as well. So the idea is you do but but the but the computation is really lightweight if you simplify it into a pure geometrical problem. We kind of do experimental driven experimental based datadriven process to understand how basically the groove contribute to the angle. We we we get a very simplified value. We do the initial guess. We put things into numerical simulation to optimize the groove pattern. Yeah. So there's like a loop later loop that's that's based on more precise physics based simulation geometric properties are dominant enough where only a somewhat linear relationship like >> Yes. Yeah. Yeah. But but but if you want very accurate result you still have to go back to more sophisticated multifysics model basically.

55:01 >> Yeah. Because this it is a complex problem right it it it's not only geometrical the material will swell as you can imagine most of the cases it gets saggy it gets like softer so the modulus changes as you can imagine will play a role the whole thing will gets thicker as well so it's not yeah yeah the thickness will change the bending stiffness so yeah it's way more complex yeah so that's something I feel you very often challenging when we deal with margin materials because none of them are like a black and white. Yeah. If you are doing a gear or like a linkage everything is so calculatable but here everything is so so nonlinear. Yeah.

55:49 such fascinating topic. It was really really inspiring to see all of the work that you have done especially with like ecology and robotics and where it all fits in. I'm really curious to hear how you see this area of soft robotics maybe being integrated into like future products that people would use in in daily life. seeing because we use like lots of really hard hardware u and so I'm just curious what the future is in your mind why >> yeah as you can see my approach of thinking about how these soft materials being used in products are quite I want to say to some extent different from the mainstream soft robotices so ours always try to go some semi-niche area yeah like a seating cedar or pasta being this is food being robots or all these at least it's one perspective I see how non-conventional robots get into products even non-conventional robotic context but but there are some mainstream applications for soft robots as well we started to see I even saw Steve showed me earlier yeah soft grapers right being mounted as an end factor for more really robotic arm and you can handle much delicate soft objects like a food or or jellyfish or whatever. and my student also started to look a little bit into how you can do a little bit more I guess intricate soft gripper. So he was working on he was showing me it's like a soft skin but you could a soft skin mounted on a rigid rigid hand robotic hand for example. you if you don't do anything it it's just rigid but then you can inflate bubbles and then becomes very compliant now you can grab a delicate cake cupcakes or whatever and then he can also do suction cup so now you can grab very smooth object I I see this is a little bit relevant to what I said about hybrid intelligence how how you leverage the smart material aspect that tune lot of physical material reality but use those to augment some of the more precise controllable machine per se. I I feel this could be one possible direction that's productable.

58:21 Yeah. Down the road. Yeah. >> You looked at the trusses examples of beds and things. >> >> where has that work gone or where is it going? It didn't so far it haven't gone anywhere because my student basically worked on the sort of the you know the the the whole concept the mesh and optimization of mesh part and we tried to propose to a hospital local hospital to do this. The proposal didn't get through so we haven't got to work on it.

58:58 We still believe in it. Actually we are right now at much better position to to work on a system like this than two years back. And we also discussed so the students thesis. so the algorithm is not only only applicable for this type of specific trust robot. so the algorithm can be generalizable for for example other closed loop graph based robot design such as tenseity robot. We actually think 10sec robot is actually more practical than the mesh because this mesh doesn't have a recoverage force to let it recover back to its undeformed state basically. Yeah.

59:37 So we likely will go that route but with those algorithms. >> Environmental DNA collection. >> So what was the this very useful interesting technology? What's what was the particular thing that that your solution addressed? What was it a particular difficulty of like quality or separating you know?

60:08 >> Yeah. Yeah. Yeah. Yeah. this is a conceptual image. it it's part of basically in the lab we we we're right now having an effort to write a vision paper. So some of the diagrams related to like eological things I just showed is part of the paper. like like those things I mentioned. Sorry, I don't know how to Yeah, like it's part of this bigger framework. Think about how you can make completely biodegradable robotic system and we've been brainstorming what are the use cases.

60:43 Environmental DNA just came into the conversation because we were thinking about you know the residues left over by animals, plants or even soil. So, we may not need a robot that does very precise navigation or anything, but we can engineer a passive system that hops or roll around semi- random, but still statistically predictable. So, we're we're thinking this could be just a way for them. Imagine Tombbo inspired system basically rolling around and collect DNA as long as we know roughly which area this robot covers. it's where we are thinking but no exact project at this moment is going on unless you have some good ideas.

61:42 Yeah. And also this right this image is showing the the fly flying machine. So this is something actually one of my students working on basically like dandelion and milkweed inspired flyers also passive. So those you could potentially use it to collect environmental DNA is in the midair versus the hoppers and the rollers will be on the ground basically. >> I have a question. So your work has so many different different source for actuation such as kinatic and a chemical reaction temperature. I'm curious like do you have a paper or what are some other potential modality of actuation that you will be excited to explore in the future?

62:32 we at this point I I feel like motor is awesome and and in terms of self actuators there's no single ideal one. So each of them has poor and calm most of them are weak and and it's really has to be application specific. Believe it or not, the wood actuator, the seed carrier indeed is one of the very stiff one because even when it's wet, the coil, the material itself has still about about five to 10 gigap. So most of the hydrogel is like a one megap or 10 megap, you know, like that's soft material. So nothing ideal. I think it really depends on the application. So far, I guess things I didn't show, but it's going on in the lab. We we through a like a a defense funding funded project we are working on biohybrid. So those would be like a real human skeletal muscle. and there are some good property of that being like selfheal even self-pacable but obviously super weak and not controllable. And we are also working I'm getting also excited about those high voltage fluid driven actuator.

63:49 So those yeah for example electro electro like a static force and all those so those type of actuators are at least electronically controllable. You pump in a voltage you drive you drive a fluid I think it's also getting closer to our vision of hybrid system but you can still make soft compliant actuators out of you know in this case hydraulic actuations.

64:20 I did ask questions before we head to SRC. >> Yeah, >> I was gonna ask one more thing in terms of ecological applications. It seems to me there's a lot of challenges in the environment for various species in terms of adapting quickly enough not going in a place that things can't adapt to, but we're going very quickly. Seems to me there might be some solutions you could add. If there's a weak point and like a plant or something where it needs like a some sort of help in delivering seeds or if there's a if there's some weak spot >> in the chain of you know the development of that species something like this could provide like >> I'm just thinking >> imagine you could provide some solution that could >> allow for a weak spot in just the time because species can evolve. in, you know, 100 years.

65:19 >> Yeah, that's a good idea. I mean, >> it might be a good funding idea at some point. I'm not sure it is right now. >> Yeah. I mean, there there, for example, sea grasses in in California, North Carolina as well. So, they change color because the weather change and they become a little bit less green to tackle I think the condition. lots of plants does that too. I can imagine some sort of color changing wearable skin for the plants maybe to to tune that. Yeah.

65:54 >> Yeah. Yeah. Yeah. Yeah. Exactly. All all those are very interesting interesting context. It's a very hard for us as non-ecologists to get into any of those. I would say it's like a pretty big learning curve. We're getting a little bit more knowledgeable about like I guess planting trees but super interest in things in ocean. Haven't got into into understanding the problem space too much yet.