It's my pleasure today to welcome
Professor Becca Kramer particularly will

visiting us from Yale University
where she's an assistant professor

of mechanical engineering and
materials science.

Rebecca I will tell you in a minute how I
came to know about Rebecca but before I

let her tell you about or exciting work
in software body aches and materials

I should say that if you are looking
to become a faculty member and looking

to become a successful faculty member you
can do no worse than getting your Ph D.

at Harvard University with Rob Wood and
then going off and

getting more awards and I could write down
so I brought my phone including the zero

in our young investigator award
the A F S Our young investigator awarded

the Forbes thirty Under thirty award for
manufacturing in two thousand and

fifteen and N.S.F. early career
award a NASA early career award and

I think that's enough that all
list without embarrassing her more

the record is clearly already a very
accomplished scientists an engineer and

I should just note
personally I heard of her

when a post aka My went off to a
conference living machines conference and

came back and said My gosh there is
a person who's doing the coolest stuff and

we have to meet them and collaborate with
that person and even invite her out for

a talk and it was Rebecca and what I
found so exciting is that the work we do

My name is Dan Goldin by the way
Seth told me at a dinner myself.

As we study how animals move
on in complex environments and

try to make robot models of those those
organisms and what we're realizing

is that we have very limited understanding
how organisms stress organisms like snakes

use skin scents and mold their bodies to
interesting and complicated virus and

this is a real gap in our understanding of
principles and I think a gap which which.

It does not permit us to make robots which
have the abilities of the organisms were

interested in so that's my personal
connection Rebecca but pleased to

welcome her day to give our i room
seminar and I'll let her take it away.

Thank you.

OK never hear me.

Thank you so much for that introduction
I really hope I can live up to the House

thank you so much for inviting me here
I'm thrilled to have the opportunity to

share my labs work with this group
this is a lovely attendance and

I have heard that there is why of range
of backgrounds in this room so I'm

hoping that I can touch on a few different
areas materials manufacturing robotics and

kind of sweeping across some interests
that might be represented in here so

today I'm going to talk specifically about
our work on robotics skins that turn into

objects into multi-functional robots but
before I dive into that I want to talk

about where we are with robots today so
mainstream robots I'm sure this being

a robotic seminar you probably see some
pictures like this very very often

automated robots manufacturing robots This
is what's typical in commercial robots

today these robots are extremely good at
what they've been designed to do they're

fast they're precise and they're very very
strong but what they're not is interacting

with their environment they're not doing
anything particularly predictable or

in unpredictable environments
I should say and

they're not working alongside
humans they're not very safe for

working alongside humans I used to say
in this particular image that there were

no humans in the photo but then over
time it was like a Where's Waldo some

people started pointing out this guy
are here and I think there's I think

there's at least one more person in this
photo so I was proved wrong on that but

there are no people working alongside the
robots so this is where we are with robots

today now there are people working
on this problem the idea of robots

interacting within unstructured
environments human created

environments are a lot of
people working on it and

that includes people here at
Georgia Tech I believe there was

a team here that participated in
the DARPA Robotics Challenge it's fairly

unfair of me to show this blooper reel
because I don't build robots like that.

But I.

I love to show this video because it
demonstrates a really challenging problem

interacting with the environment and
these robots were so

successful this is the state
of the are in robotics and

they were incredibly successful
at what they were doing but

you can see at times they did fail doing
what we consider to be simple tasks things

like turning a door knob walking up stairs
even just walking forward as I'm walking

along I don't just spontaneously fall
over or sometimes I do but that's a blip.

So I think when we imagine robots for
the future most of out there picturing

something more like this real animals
something more like ourselves things that

interact with the environment seamlessly
can switch between tatts go from

manipulation tasks to locomotion things
that switching between environments and

what there's a few of these this
one's jumping out of the water and

into the air I think the next one
hopefully is a frog that does the same

thing jumping out of the water into
the air changing environments right we

interact with the water very differently
than we interact with the hair and so

changing between these different mediums
we actually adjust our own bodies

stiffness in the way that we behave in
order to interact with the mediums it's

a really really difficult thing but it's
not something we actively think about so

it's a problem that that I've started
thinking about and how do we make robots

that can do something like this Well to
answer that question I actually actually

asked that question to Barry trimmer
toughs number of years ago and

he brought this plot to my attention
which I'm going to show you now so

this is a plot of all the known species in
the world it's pretty old at this point so

there's probably more about a million
represented here but what's interesting

about this is that every known species
every known animal in the world is at

least partially soft there are no known
animals that are completely rigid and

more even more interestingly more than
half of the known species in the world

are completely soft with no rigid
components now there is some environment.

Here are things that are completely
software typically supported by their

environment so they're either
in the water like the octopus or

squid fish will fish have
a skeletal structure but

are very very soft more soft than we are
also things like worms that are supported

by their environment when they're
in the dirt are completely soft but

this is really something clicked
in me when I saw this for

the first time because I thought if every
known species to least partially soft and

over half the known species are completely
stopped maybe material actually

has something to do with our ability to
interact with our environment maybe it's

something as simple as just material
properties so I became really interested

in soccer about so I'm going to go
through some of the advantages of soft

robots what I believe are the striking
features of software what's moving forward

before I talk a little bit specifically
about our work so these are a few examples

I love this one this is the mesh worm
This was originally developed by a song

make him when he was a post-doc in my
lab at Harvard we overlapped a bit and

I worked on this a little bit with him and
then he's now a faculty MIT you think this

is a really simple system there's
an internal balloon there is a finger trap

toys a memory alley actuators just wrapped
around it causing peristaltic locomotion

and you can see hitting it with a mallet
and it just keeps on doing its thing.

Another pretty famous
example is this iteration

of the softer robot there was
a smaller version of this that didn't

that was tethered didn't have any on board
Air Supply this one is untethered you

can see there a car rolls over it you
could certainly comment that the car

rolled over just the soft legs and not the
internal structure but in general I think

it speaks to the resilience of soft robots
that they're resilient to impact and

stretching stressing
mission environments so

other benefits of soft robots
they can change shape and

they can change their size potentially
access denied spaces you hear a little bit

of military speak there and this is a this
is something that I think a lot of people

are becoming really interested in more
figure robots now the video showing here

is not a robot Unfortunately this is
the real octopus trapped in a bottle and

you can see it coming out of the bottle
but you can see at the length scales

of this animal are larger than that
bottle whole and it's able to squeeze

itself through it's just amazing
what it's capable of doing and

if you actually if you go on youtube and
just look up octopus moving through to

there's some really amazing videos of
this of the octopus going through these

long complex two patterns and the way
they can change their body side and

manipulate themselves is just
unbelievable so major motivation for

wanting to create robots that
could potentially do these things.

Soft robots could potentially be
made safe for human interaction and

handling diverse and fragile objects so
the video here is highlighting that second

point handling fragile objects this is a
robot made by softer by the sink based out

of Cambridge Massachusetts and here it's
using a pneumatically inflated gripper

to grasp very fragile and cooked eggs you
can see it's very successful at doing so

in the image below you see a grasping
tomatoes and this is actually huge market

for softer radix now using these soft
grippers to manipulate fruits and

vegetables without bruising them
huge huge market into agriculture.

They also have the potential for
interfacing with biological tissues and

wearables there's a lot
of examples of these the.

Types of examples like this these days but
the want to highlight here is one from

Connor Walsh's group at Harvard where
they've put sensors all over a human and

they've actually coordinated those with
actuators and they've been really really

successful at augmenting load capacity at
actually increasing the efficiency and

reducing energy output from the wearer
by coordinating these sensors and

actually has to fire the actuators
at just the right time so

it's been been really successful
interesting work to follow.

And I lump these two last categories
together soft robots could potentially

be very cheap to produce and
they can also be extremely lightweight and

I really attribute this again to the
materials that soft robots are potentially

made out of very soft to last summer's and
phones very lightweight materials so

when you imagine trying to make a typical
manufacturing robot like we saw

the automated automotive manufacturing
robots in the first slide

they're very very heavy and
they're all metallic So

they're certainly not lightweight here
we can see the video played again.

So you can see this is the activity that
came out of Jennifer Lewis's group I think

about two years ago hopefully that's
accurate where they've completely

printed an autonomous robot
the first fully autonomous robot and

it doesn't actually locomotive very well
it just kind of moves its limbs up and

down but
it's completely materially controlled and

they use this really really unique fluidic
control system imbedded with it and

the thing is entirely printed so I think
it opens up a lot of opportunities for

soft robots could go so
with the benefits in mind I hope

that I can at least rely that you
see the advantages of a softer so

with the feasibility of Dr Watson
the benefits of soft robots just accepted

I started to think about a problem
within the softer about expats and

that is multi-function ality So all of
the robots I showed you were designed for

a very specific purpose this locomotion
robot was designed to look about and

you could see it being hit but it was so
local noting the manipulation the gripper

you saw was designed specifically to grip
an object but it wouldn't be very good for

locomotion the wearables are designed to
be wearable but they can't locomotion

their own they can't grasp objects on
their own so none of these things can go

between these different functions without
a complete redesign of the fabrication So

the question I wanted to ask is Is there
a more flexible design approach for

softer robots can we make something that
can actually do multiple tasks like week

so all the interact with
the interacting with the environment

something that can look about that can
manipulate potentially be used for

wearables and accomplish all of
these tasks using the same hardware.

So when I was asking this question this
was several years ago and around the same

time there was a solicitation from NASA so
Dan mentioned NASA E.C.F.

award and there was this solicitation
saying hey we're interested in soft

robotics technology how can soft robotics
potentially benefit space missions and

space applications and so
I started thinking about some of

these same questions how can we repurpose
hardware you know if specifically for

space applications it's incredibly
expensive to send things up things that

are heavy things that are bulky are very
very expensive so if you need to perform

multiple tasks of an unknown environment
you need to make an exploratory locomotion

robot that can go out and collect
data from an unknown environment and

then you want a manipulation robot
continue manipulator maybe they

can collect samples from that environment
if you need suits to help promote blood.

Within the astronauts that are out in
space you would have to send specific

hardware for each of those purposes
it's very costly so I started to

apply the same question to specific NASA
contacts can we create repurpose for

hardware that can do all of these
functions and the idea that I had

was robotic skin and so the concept
of a robotic skin is that it's a skin

like planar conformable material with
embedded actually action and something and

we can wrap it around inert to form bodies
to impart motion onto those bodies so

we can take the skin we can wrap it around
something like foam or an inflatable and

perform some task and then reoriented
to pick a different motion perform

a different task even take it off put it
on a totally different body to go for

a different task so in the in
the cartoon you see here this is an idea

of a robotic skin that has maybe
linear actuators that contract and

we wrap it around this balloon which kind
of looks like a good one orientation and

it really contracts you can
imagine putting links of these

together to get something
like peristaltic locomotion

then we can take that exact same skin take
it off turn it ninety degrees put it back

on now the actuators are oriented
along the length of the body and

we can track them we get bending motion
and you can imagine putting several links

of these together we get something
like ensured locomotion so

we can get different modes of action we
can get different modes different gaits

different forms of locomotion
using the same hardware

just reconfigured on the surface of a soft
body so that's getting this point here

you can turn any deformable in animal
object into a robot I pull a balloon out

I wrap a robotic skin around it turn it
into a robot also because we're doing

everything from the surface we can
measure state really well so it's as

opposed to embedding sensors within a body
we do everything from the surface and

finally we can utilize additive and
text how many

factoring approaches to make these things
the fact that we make everything in two D.

really opens up the manufacturing space so

now we can use printing
technologies put everything into T.

and worry about conforming it to three D..

So to start I'm going to focus on
the second two points about fencing and

manufacturing approaches for the fences.

So right when this project started we
happened to be working with liquid metal

based resistance sensors so the idea is
really simple we take a conductive fluid

and we put it in my car channels
within a thin elastomer sheet and

then by straining our pushing
are bending the micro channels

it changes the resistance across the
channels and we can map to the defamation

the particular liquid metal we were
using is gallium Indian alloy for

those of you not familiar with this it's
a wonderful very interesting material as

you can see in the image here this is
an image taken by Michael Dickey and

state and it's so galley even you alloy

it has about a third the kind of to be
a copper it's liquid at room temperature.

And sorry this is someone else's
pointers I'm still getting used to it

you can see here that it has these
really interesting kind of solid

like properties even
though it's a liquid so

Gallia oxidize is very readily in any
option containing environment so gallium.

You put it into this environment in the
gallium oxidizes and it creates this oxide

that allows you to print three
dimensional structures using a liquid

metal onto a surface so here this is
three dimensional it's holding itself

very very well but if you were to just tap
on it put any kind of pressure on it or

even shake the table the whole
thing would collapse into a puddle

very very interesting material
to work with lots of.

But extremely hard to manufacture with so
it's kind of like trying to

print with honey even though it doesn't
have the same properties it's very viscous

in this case we have an artificially
high surface tension induced by this

Galleon oxide formation on the surface so
the combination of the viscosity and

surface tension density of the fluid
makes it non-printable and

it's made of condition so
we had to an idea to try to break it up so

we did as we've put the liquid metal into
a vial we filled the file with a care.

You're solvent so

we used to tell you in ethanol water this
could be many many different things and

we just applied a ton of energy with the
tips on a caterer and we broke it up into

these nano particles that are suspended
in the carrier cell bit and this is now

principle because the printing properties
take on those of the carrier solvent but

it's carrying all these liquid metal nano
particles with it and each particle is

isolated from the others because they're
all coded in this gallium oxide skin so

I don't recall a spontaneously so
interestingly this works so

you can see printing onto a nitrile gloves
and if we zoom in on this you can see our

beautiful nanoparticles on the surface
we put down droplets of the carrier

solvent containing the nano particles
to carry solvent evaporates away and

we're left with films of these nano
particles on the surface now for

the same reason that they
don't spontaneously coalesce

the is also don't create automatically
conductive films the gallium oxide isn't

very conductive very noncommittal slightly
conductive but there's too many electrical

losses between all these particles
given the semiconductor Vittie of

the oxide we just don't see any resistance
through the through the film so

we have to do is apply pressure we found
that if again we just tap on the surface

it'll break all of these open so
they're kind of like balloons carrying

the conductive material on the inside and
we just break each balloon and

they flow together and
create these nice conductive channels.

So you can see the glove
that we created but

unfortunately as much as this seems
beautiful this one didn't actually work.

So we printed string gauges onto
it we have contact pads here and

we can get any kind activity out of
it now it turns out there's a couple

of reasons this is actually on
the cover of advance material.

And so people often ask
about the glove and I have

to admit it didn't actually work it just
showed that we could print the material.

But there's a couple of reasons for why
it didn't work and one of those reasons.

Has to do with the fluid dynamics
of the printing process so.

I'm going to make you interact with this
in the second here but we it turns out so

printing into printing in particular is a
drop in the man process meaning if I want

to make a lot I can all of you in the back
kind of see me because I'm actually great

if I want to make a line a printed
line and I put a drop here and

then I put a drop here and I go back and
put a drop here in put it up here so

it's not it's not a linear process you
don't put a drop here and then keep doing

it in a row you actually go around this is
called Print mass design that has to do

with how fast the carrier solvent is
evaporating away you want to drop here and

then you want the carrier solvent to
completely evaporate before you put

an adjacent drop next to it so that the
particles don't migrate into the next drop

so it makes sense yeah so we started
looking at this and we said OK we

want our sensors to have this geometry we
want them to be beautiful rectangles but

unfortunately when we look at them from
across section we see geometry like this

and we see geometries like this and then
we started looking at individual droplets

and how the particles are settling on the
surface and we realise that depending on

the carrier cell that we were using and
the surface we're printing on

the particles were all aggregating on
the outside or piling into the middle and

I know why this would happen
surface tension is part of it

with my grand attempt to
interact with my audience.

Yeah they hear men going.

Yet that's exactly right OK so that's
even more specific than I was looking for

as they say internal droplet flow the yeah
so this I don't know about these prior

to doing this process there's some fluid
dynamics person in the room that's great.

So it turns out there's these internal
droplets flows when you deposit drops onto

a surface you know they're so small we
put these drops out on the surface and

depending on again the carrier solvent and
the substrate we were on

it was either concentration driven or
thermally driven and that would derive

that would drive the direction of these
internal droplet flows called meringue.

So in this case we would have meringue
only flows going to the outside

all the particles would just
settle the outside in this case

we would have meringue on inflows
forcing them in words and

they would all pile in the inside either
way we are ending up with these printed

traces that had just clumps of particles
that weren't really well connected and

things were not working as we were
trying to print these devices.

So I had a very clever post-doc at
the time well bully who is now a faculty

member at Boston University and he
introduced this concept of CO solvent so

we mixed water and ethanol together so
previously we had only been considering

the carrier solvent as a single salt
it would be like water tell you in

here we mixed to have different
evaporation rates and something really

interesting happens when you do this you
put the drop down on the surface and

the two solvents segregate from each other
so in this case ethanol evaporates faster

than water so when we put the drop down
in the surface ethanol comes out to

the surface of the droplet water stays in
the middle of a water in which core and

then we coated all of our liquid metal
particles with us hydrophobic so

we wanted to follow the ethanol they
didn't want to be in the water so

after we put the drop down all
the particles would go up to the surface

get trapped by surface tension and
if we are clever with our concentrations

of how many particles are in each
drop they would all come up to

the surface congregate there and as a drop
of operates just settle down nicely.

Into this beautiful uniform film.

And this turned out to work really well.

So we were able to make nice model layers
and multiple stacked layers of particles

they interestingly worked really well as
mares and filters and they also worked

really well as principal sensors so
this was the follow on of that work we

were able to print very very nice uniform
layers of liquid metal and then we wanted

to make an all printable process really
leveraging this idea that we can make.

Me in two D.

and then wrap them around three D.

objects later we wanted to really
dive into this all principle idea so

we made our printer first
print the particles and

then we put down a little bit of.

Elastomer and let it completely cure there
so it had this kind of a last summer

tip on it and then we use that to tap out
the conductive pattern that we wanted to

actually isolate so we were able to fully
print the last really on the bottom fully

print the conduct of particles over it
top out the pattern that we want and

then encapsulate the pattern
after we had tapped it out so

here after we printed the last more layer
on the top it was really nice because

the elastomer would actually filter in
through the particles and make it so

that they could never be coalesced so
the particles that were not coalesced by

the printer were permanently That way the
particles that had been tapped out in this

pattern were permanently that way and
these sensors were quite successful so

we made a few different types of sensors
out of these this is a pressure sensor

you can see so just putting light pressure
on the surface changes the resistance.

And here we are able to make long
conductors that were much larger

than the scale of the printer that we were
actually printing on so we had a stage

that was about this big but by making
the spiral patterns we could make really

really long conductors by stretching them
out and then finally we were able to

get really intricate patterns so here you
can see Sponge Bob and as an interest.

This paper was published in advance
materials if you actually go and

look up this paper you won't find this
image in there we tried to submit it but

after accepting the advance
Carol came back and

said that we didn't have copyright
permission to use the Sponge Bob image.

So I e-mailed the colonia and
asked if we could have permission to print

in a technical publication
I never got a response.

And so if you actually go look up this
paper you'll see an image of a world map

instead.

OK And then to really round out this
work most recently we've been looking

at Laser centering so you probably saw
from the video just there that tapping out

the pattern and applying this mechanical
centering process is quite slow and

we wanted to speed this up get really
small features and make it more precise so

we introduced laser centering we now
just put these printed substrates

in a laser and we can get really more
complex intricate patterns as is this is

the fabric Tori gear my lab is called the
fabric in this is our lab logo you can see

the nonsense turn and centered patterns
in a more intricate pattern here with

medical it he said just applied to the box
that we can use it like a wearable and

we've been able to create
really really small things so

this was actually centered
in focus I am beam enough.

So there's no thermal effect here
we're trying to isolate the relative

effect of a thermal effects of the laser.

And then we can also make a multi-path
turned things so if we take a film

of the liquid metal nano particles we
center the surface then there's some

non-conductive layer underneath there and
so we can actually put another print

another layer of the particles and
center again and create isolated or

preferentially connected electrical
circuits multilayered circuits so

this paper just came out and A.C.S. a five
inch interface faces a couple weeks ago.

OK So we've done a deep dive
into liquid metal particles and

using them as sensors and trying to apply.

Printing techniques this
robotic skin concept so

I'm going to take us back now and
actually talk about how we use this

how we use these devices as sensors in the
robotics team concept so we're going to

start with just sensory scans and
ignore actuation for a little bit longer

Here's some first iteration of our sensory
skins made with the liquid metal So

here is just a triangular module we
have liquid metal strain sensors along

each edge of the module we pull
the nodes each of these stretches and

we're able to reconstruct what the module
looks like we can patch many of these

together we can reconstruct state insert
information across an entire surface and

one of the things that I really love about
these particular sensors is that they're

independent of the material properties
that they're actually in cased with so

typically resistive sensors if any of you
work with conductive composites you have

a big problem because the electrical
properties are tied to the mechanical

properties of the elastic or here because
we have liquid metal encased in elastomer

there's actually a difference
between those two things.

So to prove this we were looking at
stress relaxation in particular so

we took the sensors and we strained them
and then we just left them strange for

a really long period of time and
here we see this really standard stress

relaxation occurring in the material
the elastomer That's in case what then but

the sensory signal stays the same because
that the same length the entire time so

it really showing isolation of the
electrical properties from the mechanical

properties of the encasing material and
a real advantage of working with this

particular sensor technology now one
of the disadvantages however comes back

to that gallium oxide keeps coming back to
haunt us so we in case these things and

we strain them and then we put them on a
robot and we string them over and over and

over again and we were losing state
information and we're wondering why so

we looked at this in more detail we
ran it over hundreds of cycles and

we were able to see a significant
signal draft OK So what's causing this

just stressing over time the gauge factor
is drifting makes it really hard for

us to actually use these sensors and
input.

In Taishan Like I said it comes
back to that gallium oxide So

it turns out what's happening is every
time we strain the sensor the oxide skin

in case liquid liquid metal breaks and
new oxide forms and then we release

the sensor and that new oxide folds back
into the material then we started again

same thing happens the skin breaks and you
oxide forms release it it goes back into

the material so with every single
strand cycle we're introducing more and

more oxide into the system and as I noted
earlier the oxide is non-conductive so

reducing the conductivity of the sensor
every single time we use it unless we're

simultaneously tracking the length and
calculating how much oxide should be

forming with every single cycle we're
not able to actually track this drift so

it makes it really hard to use
these sensors even though they

have great benefits and we've made
them a principle we've struggled to

actually implement them in the robotics
concept so we turned to a different type

of sensor technology these are conductive
composite base capacitive sensors

there are lots of conductive composites
out there any elastomer with a conductive

filler will act as a nice elastic
conductor here we put them in a capacitive

layout so we have our top electrode
dielectric in the middle and

a bottom electrode standard passively
out we can track past events

the particular conductive filler we used.

In exposed graphite So
here you see a really hot crucible and

we just take our graphite and
put it in there and you'll be able to

see it start popping so what we've done is
we've just taken flakes of graphite and

we've soaked them in a solvent and
then we put them in the crucible and

the solvent boils which forces
the flakes to pop out like popcorn so

this is a graphene but
they're very very thin flakes of graphite

typically referred to as exposed graphite
and I think it's also pretty fun to watch.

So these sensors turn out to work
really well in implementation

they have this nice linear electrical
response we see a standard standard

nonlinear mechanical response but again
we expect that from any elastomer and

we were testing them up to one hundred
thousand cycles and seeing the Pentagon

how they are manufactured we see very
very little drift in the gauge factor so

these three colors represent
different manufacturing processes and

we ultimately selected the film based
process which is just Rod coating

due to this really almost no gauge factor
just over one hundred thousand cycles and

interestingly when we first did this
work we actually tested I believe to ten

thousand cycles which was the most
that we had seen in the literature but

then reviewers came back and
said it wasn't enough so we we went for

one hundred thousand which now I'm fairly
confident so correct me if I'm wrong but

I'm fairly confident that this is the mess
that something is the president military.

And we also looked at when they would
fail we found at least two hundred

fifty percent strains sometimes more again
depending on the manufacturing process and

we tested them in a variety of
temperature thermal conditions and

found that they were quite stable and
between twenty two and fifty

degrees Celsius we're currently testing
them in different humidity conditions and

currently finding that they're
stable across those things too so

OK we've identified a pretty
good stretch sensor

an elastic sensor that seems to work
well that's kind of part of our That and

one way that we did that is by putting
them in some robots so we started this

collaboration with other lab based out of
San Francisco they create these really fun

robots these inflatable fabric based
softer robots they make them at different

scales you can see some human scale ones
and when they bring them to conferences in

conventions it's actually really fun and
they allow you to operate and

you can have a little robot fights using
these things and here you can see.

A group are attached to an arm here that
they've created and they came to us.

At the right time several years ago right
when we were kind of working on these

sensors saying you know we need to control
these robots better we have no way of

controlling them because they're
soft we can't use standard sensors

let's try to put
the sensors on the robot so

we started to do exactly that here you
can see a one degree of freedom arm and

one of our sensors is placed over that
arm and then this is the next iteration

of that where the arm is underneath this
white sensory sleeve we called it so

there's a couple of different sensors on
the sleeve there are sensors in the joints

both on top and below and then we also
have and I am you touched on here and

motion capture markers that we could
validate the sensory data that we were

getting and you can see some of
the hardware like the signal conditioning

boards colocated with each sensor and
these sensors were printed directly into

the fabric so these were extrusion
printed directly into the fabric and

to create this fabric sensory sleeve and
you can see that it's a great matching So

we track the height of the end effector
and we track the angle of the joy and

you see great matching between both
the I.M.U. and the motion capture system

which validates our sensors are giving
us accurate state information and

the reason we track both of these
characteristics both of the angle and

the height of the end effector
is because the system is soft so

in a typical rigid so Sam if you
are checking the joint angle you should be

able to back out the head of the defector
this is moving here so this has to

be a certain height because
it's a rigid link between them

in our case we didn't see that especially
when we loaded the system up so

you can see this bar sticking out of the
end oftentimes we would put out the arm

and then put weights at the end of it so
it was actually holding a lot of weight at

the end and
then when we tried to bend the arm or

may have the joint emboldening we would
see buckling in interesting places so

we would see buckling at this joint
here and also something that we

were referring to as the S.

curve where the arm would
actually drop down at the elbow

some kind of shape in the middle due
to the compliance of the materials.

The joints and so we do have to check
those two things separately but

you can see that we were able to do so

with just the sensors that we had
printed into the sleeve OK So

we wanted to make the system a little bit
more complicated so we started using.

Let me move as we can here
roughly a little bit OK so

we started looking at a gripper also
made by other lab this is a three finger

you can see it grasping some objects and

one of the interesting things about
this is that it's packed in the ploy.

So we can deflate it package or to take
up a really really small volume and

then deploy it when necessary this
is again another NASA funded idea

project where we're looking at taking up
very little space during transport and

then deploying when we get to a location.

So we created some three sleeve began.

Here you can see our sensors on the sleeve
we have a long sensor that's going to

be located on the outside of each
finger and we have pressure sensors

located the tip of each finger in the
middle of each finger in order to identify

what we're grasping and
we just put those right over each finger.

So in this video you can actually see
it doing some grasping tasks that in

a minute you'll see a pattern deploy
a task here it's grasping a water bottle

We're tracking both
the position of the fingers and

the gripping force as it's
holding that water bottle.

And here the system is fully packed
down with the sensors on board so no

red Velcro parts that's where we actually
place the sensor we have a little bit of

flexibility with where the sensor is
placed on the finger by just building it

on when the thing is packed down you see
that the data is completely nonsensical so

there are the sensors are slack we're
not getting any information out of them

as soon as they deploy up we
see nice state information and

then they open up completely we continue
to be able to track that state information

without any manual intervention.

And so we did some close control with
this here we have close open close open

step signals and then you can see a little
bit more intricate steps they will down

here you can also see some offset so the
blue is the target the setpoint in the red

is the actual data that we're collecting
we attribute this offset to the sensors

actually being non optimally placed on
the fingers and maybe even moving around

a little bit during the deployment and
so to attack this we started looking at

design for control how to actually play
sensors on a body in order to make them

give us better tracking
a set point a lot better so

this is a recent paper that's actually
going to appear in a few weeks and

this is just by directional bending
actuator So it just goes one way and

then the other and it has the sensors
built directly into it so

the sensor is is leveraging a fabric layer
those both the strain limiting layer of

both actually others in order to
enable the spending back and forth and

then using our conductive composite
to create the other electrode for

the capacitive sensor we're
able to create sensors and

strain limiting layers that are one in
the same and built into the system and

we are able using just a standard
controller nothing fancy going on with.

The control we're able to get really
really nice matching that point so

this tells us that placement of
the sensors within the system

is actually quite important important and
that building sensors and

actuators and bodies and trying to
put them together as an afterthought

is maybe not the most advantageous
way to design these types of robots

rather we would want to think about
control as we're designing the robot and

co-locate them as much as possible
to look at the components.

OK so now that we have fairly
good control we've proven.

We can use our sensors for control and
they're viable within planar things

we can move to actually
tracking actuation and

co-locating actuators and sensors within
the same robotic skin to actually achieve

this vision of robotic scans that
impart motion onto deformable bodies so

here are some of the first
prototypes that we've created.

We keep the robotic skin concept
pretty broad The idea is that any

skin like substrate any plainer substrate
that has invested actually eaten and

sensing with and it can be considered
a robotic scan that's not tied to

any one specific type of actuator and Eve
type of sensor any type of substrate or

any way that those components
are laid out relative to each other.

So the first one that we have here
is a it's a fabric skin with printed

conductive composites capacitive sensors
into it and then shaped memory alloy

just ignore the cloud of wires
coming off of it then you can see

a smaller module of this again the sensors
are printed directly into that fabric and

you can see the nice shape memory Ali
coils that are associated with it and

here we have another type of robotic
skin that's alas are based so

it's using an elastomer substrate
with McKibben style actuators

in the underneath each of those actually
others is our sensors again the capacitive

sensors that we've created in
order to track system state.

And so we have this general idea that we
can create them in different different

shapes so we have square ones that we've
created with with parallel components and

those accommodate singular
curvature bodies really well.

Things like cylinders tubes of foam so
we take the skin and

we wrap it around a tube of foam by
contracting the actually matters it

allows the tube to compress if we then
take the same thing reorient it allows

the tube to Ben and then we can place them
together to create more complex robots

like this three segment continuum and if
you later we also created this triangular

robotic scan which accommodates compound
curvature so if we want to put something

want to put the skin to something like
a ball we would need to have the actuators

not just parallel to each other like
you can see in this illustration and

then here one of the examples we did
was creating twenty of these skins and

putting them on the faces of
a six part time secretary and

then using them to make
the ten seventy ball roll.

So I just want to validate that we
were able to use our sensors for

closed loop control of these systems here
this is showing a skin that is controlling

a single segment using a pneumatic
McKibbin style Actually it or this one's

using a shape memory alloy actually enter
but both regardless of the actuation

technology or the substrate material are
able to track a set point extremely well.

And one of the I think more interesting
things that we've introduced recently this

actually was presented this past April
was that we can use the sensors and

actuators in the skin themselves to derive
materials about what they're wrapped

around so we don't even need to know
anything about what we're wrapping around

we can take the skin in a very simple case
if we know something about the morphology

of what we're wrapping around so here
we're assuming that we know that we're

wrapping around the cylinder but that's
the only real assumption that we make and

then we can take it around wrap it around
the cylinder produce some motions so

use the actuators either compress
the cylinder or to bend the cylinder

track what the actuator force is
actually producing with our sensors and

then use that information to derive the
material properties of what we put it on

this is actually pretty straight
forward so we can derive the bending

modulus pretty well just knowing the force
of the actuators again the radius

that we've wrapped around
the curvature that we've produced and

then assuming linear properties which
is valid up to about fifty percent and

for most of the definitions that we're
producing that's well within that range

we can easily extrapolate
over to Young's modulus So

dividing the material properties of what
we've wrapped around without actually

knowing that in advance and then the skin
can then build up its own control model so

we can take it off of something put
it onto something else have a derive

the material properties and
control model to control that thing and

keep doing that without having to know
anything about the body in advance.

So with that I'm going to start showing
you some of the demonstrations we've

created using this robot
it's going technology so

I showed you this before this
see this continuum road.

Lot It has three robotic scans
wrapped around the cylinder of foam so

each robotic skin controls
motion in one part of this and

then we take those exact
same skins remove them and

put them on different things in order to
get three forms of locomotion So this is

a skin that's Lokomotiv on its own we call
it the body you body because in short

without a body here we have a skin wrapped
around foam with these A last summer and

perhaps just pinned in you can see
the pins just taking out the size this is

really just designed on the fly and
it's able to do a rowing locomotion so

it throws these feet forward pulls itself
through the fifo or hold itself and

then here we have a motion
with an actual body so

you can see the skin wrapped around
the cylinder and these feet just kind

of using an elastic really just slapped
onto their i don't look at modes forward.

I don't know if there's
sound in here is there not.

You know I hold on.

This dancing robot really
needs need to sound.

It's not.

That's OK you can just
imagine it's dancing.

It's a short video don't worry about it.

Yeah yeah it's really OK
don't worry about it I think.

This is yeah this is the video of what
I just described the dancing road.

You know we take off the scale wrap
them around something else and

create different locomotion robots and if
you're interested in seeing the video with

sound this is actually a supplemental
video that's included with

a publication that's coming out in Science
robotics actually a week from today.

So that we can take these scans and

apply them to a number of different tasks
so you just saw the Continuum robot

you saw locomotion robots here we
have a continuum manipulator So

again three segments making a continuum
arm and then we just have two plates and

we attach the robotic skin in between them
and we use it to just pinch an object and

be able to move that object we created
another type of grouper as well.

Showing an active figure in a passive
finger so this is the passive one and

here there's another cylinder
just put on there and

we're able to create two
types of papers on the fly.

We show one application of
a locomotion robot This one has light

sensors that are attached to the front and
it's programmed to follow the light so

you can see it there trying
to follow that flashlight.

And in just a second it takes
a pretty impressive turn.

For what it is.

So one of the things that we looked at is
comparing these systems to purpose built

robots so purpose built locomotion robots
in particular that have been demonstrated

in the literature because if you're
building something on the fly the idea

is that you can take these scans put them
on something perform a task take them off

put them on something else perform
another task probably the task if if.

Isn't going to be as high as if you
made a robot that was specifically for

that task but we actually show we're
comparing two purpose built robots

the literature that we get comparable
locomotion speeds especially for

the locomotion robots OK So
this is this is one of our favorite Dimas.

So this obviously is a stuffed animal.

This works really well when we have
tour groups coming into the lab

we actually have a box of
stuffed animals in the lab and

when targets come in we'll let them choose
one of the stuffed animals and then we'll

throw the robotic skins on and robot
a size that stuffed animal on the fly.

So you can see our little horse stuffed
animal making its way across the bench.

So these are all open loop actually.

Right now and they're just they're
programmed to just secretly.

And you can see that
the whole thing is tethered.

That is your question yes you know it's
pneumatic is that the question yes

that this is this is using the Kevin
sounding matting Actually I thought

you were asking about the controller.

OK yeah so there's there's are.

They are underneath and

you can't see them because they're covered
at the top layer but there's four McKibbin

actuators so do you remember the parallel
component square the skin so

that's exactly what's wrapped around
each of these and they're all oriented.

Along the length of the leg so therefore
actually there is in each of these skins

it's wrapped around a single leg sixteen
actually turns on the entire robot.

OK And here's our trusty Ted thank
you both so I mentioned the six

bartend secretary for those of you who
aren't familiar with ten secretaries.

The structure is made of strengths and
cables or elastic segments and

if any structure where the struts
don't actually touch each other and

they're completely suspended
by the elastic segments

in our case elastic bands here so
we did is we created a six part.

Which is a popular time secretary
can figure in robotics and

there's a larger model of
this being developed by NASA

they call it Super Bowl bot it's
huge it's about as tall as I am and

they actually made it by changing the
lengths of these elastic cables in between

the segments here we wanted to ask the
question can we do this completely from

the surface can we actually attend Segarty
just using membranes placed on the surface

so here we made twenty different robotic
skins and wrap them around this ten

segregate ball and
as you will see we do produce locomotion.

It's a little underwhelming but.

It's happening.

And then we take these exact same
triangular robotic scans and use them

in a wearable application so this is my
postdoc Joran and we've just put double

sided tape on these scans and slap them on
his back we have sensors are located along

his spine that are tracking his posture so
every time he passes a threshold

the actuators pulse to communicate
with him that he should sit back up.

So after wearing this for

a very long time he confirmed that
this is indeed highly annoying.

And that he felt the need to sit back
up every time I am the system told him.

So you can see the sensory data that we're
collecting and passes that threshold

pulses the actuators and you can see that
they're quite plain or two from a profile.

They really are not too apparent So
hopefully I've convinced you

of the concept of robotic scans on the
multi-function ality that they can provide

the idea of creating hardware that can
be applied to inert soft bodies to

robot assize them turn inanimate objects
into robots and before you can clear it

I wanted to show you some of the most
recent work that we've done with this.

We submitted in our paper earlier this
week to lots of you do that as well

as a robotics deadline.

Trying to translate this robotic skin
idea not just from the formal bodies but

to moldable bodies so we wrap it around
something like foam we bend the foam and

it just goes right back if we let go but
if we wrap it around something like clay

we can actually use it as an active
sculptor from the surface so

this was our first
instantiation of this concept.

This is a a bunch of clay in the middle
and just like a real sculptor just like

an artist would sculpt clay into different
shapes we can apply surface stresses and

strains and pressures in order to actually
remove the internal clay and put it into

a different shape So here we've wrapped
we're wrapping two different skins This

is also showing the layering of the skins
the skin that you saw wrapped around there

initially has pneumatic bladders that
are meant to Lokomotiv the system so

after it's wrapped these pneumatic
bladders will inflate and

it can locomotives and
this other skin is able driven and

is meant to actually more the robot so
here it is an application

we have this thing rolling along the
surface using the locomotion skin that's

under the more things again it comes
into an obstacle and really struggles.

And then here you can see how we're
able to use the skin to change the shape

of the robot so
it can overcome this obstacle.

MERIK.

And then it's able to roll away so

this is this is of course demonstration of
this concept but it is proof of concept

that we can potentially use the scans
on multiple bodies to drastically

change the morphology of what it's wrapped
around and of course this is the vision

we are quite far from that we can wrap
the skin around a ball of clay and

then morph the body on demand
using just surface actuation

in order to accommodate different tasks or
accommodate different environments and

we just received a new effort
in order to pursue this idea so

I'm going to shamelessly advocate that
if any of you are looking for post-doc

positions we are currently hiring in this
particular area so if you want to come and

build sculpting robots morphing robots
based on robotic skins working clay

definitely contact me so with that I'm
going to and I want to thank my students

and post-docs that did lots of this work
in particular a lot of the liquid metal

work was done by Shannon and a lot of the
robotic skin work was done by Joran booth.

Here and I particularly want to thank NASA
who funded the Met majority of the robotic

scans work under my NASA and
I want to thank you for your attention.

The book with.

Both of.

You back with.

So the sensors connected composite sensors
are fabricated but there's several

ways we can print them directly into
fabrics and we have to do that for using

a fabric substrate the best performing
sensors that we've made are made.

Rock cuttings so we ride.

And then we'll basically ride.

Half the center and fully half so
that it's a match and they were able.

And they were able to laser
cut out the sensor shape.

Yes.

Yes.

I recognize the problem it's a really
good question we don't have specific

plans to look at that right now
I'm I've been following the literature

on principle planar batteries and
energy sources.

So most of the robots you saw in the demos
the working ones were primarily using

the Kevin style pneumatic actuators so for
those of you who work with these types of

actuators there they have a lot of
advantages they're strong and really fast

and so that makes for a nice you know good
demos because they work at a good time

scale things like shade memory alloy are
advantageous because they're electrically

stimulated so that's a lot easier to
put an energy source on board but

we have issues with how
slow they are actually in.

There's a host of issues with different
types of actually there is so

your question about having an energy
having energy storage on board is

highly tied to actuate or choice and so
I think that that's kind of a parallel

thing that we need to continue looking at
is how do we choose the right Actually

it is that that will do the task that
we're interested in and have compatible

preferred electrical stimulus so that we
can have onboard power so your question.

On the.

Books.

Will.

So we're we're not using any
printed circuit boards here we

do have some commercial circuit
boards in here so you might have seen

the signal processing boards that
are colocated with each sensor.

We don't have very strong processing
requirements so I mean really we

just have the signal processing components
and we keep those really really tiny and

then we're just looking at our strain
sensors and those are the only other

conductive elements distributed throughout
So if we were to integrate the controller

if we needed if we're going to
integrate the controller and

power supply then I think we would see
more need to potentially put in P.C.'s but

those are usually fabricated on caps
on things that are collectable but

not stretchable and depending on
the application in a lot of cases were

really relying on the last to City of
these scans the sensors are all pretty

strained right so the sensor goes slack
it's not giving us any information so

the sensors are put on
there pretty strained and

they're relying on the strain of the
substrate that they're on or embedded in

to give us state information so not having
that stretch ability and to the skin

itself would really constrain how we
could use the system so as your question.

Goes from.

There.

Well.

Joy.

Or.

Yeah we explored that a little bit so

we did make an upper body posture shirt
this was in collaboration with Kathleen C.

and go at the University of Michigan
she helped us with the user studies so

we fabricated several upper body posture
shirts with distributed sensors and

then we had students at the University
of Michigan where that and

we collected data we did motion capture
simultaneously to see how well our posture

sure was giving the state data and
pose data.

In comparison to the motion referenced
in the answer it was quite good so

we wrote up a great script on that and
it's under review right now but

there's the same issues with well there's
issues with all of the different options

with motion capture you have it's
static rates in one place you

have cameras that are positioned
it's hard to transport that anywhere

with our posture sure we still have
the colocated stable conditioning boards

we have some rigid electronics that the
wear has to have on them and it needs to

communicate with with something that's
processing with sensory information and

we were just wiring using communication
protocol usually eye to see.

But these are still issues that are a
little bit uncomfortable for the wear.

Yes we do we we have we
definitely have shown that.

Little more than.

One.

To print an actuator I would love to see.

So there are certainly people doing this
right there's people who print actuators

based on hydrogels For example I think
it would be possible you know the same

way we can print the passive sensors
we could print elastomer actuators.

But we have to get it comes back
to actually need where you have

a strong need for force we have a time
scale that we want to operate on.

And the sensors are that the actually
there is that we have found to be

printable don't make for very compelling
robots quite yet so it's something

that we're actively interested in but it's
a good point we focus a lot particularly

the sensors because that's within
something that we can do the actuators

we typically fabricate separately and
integrate into the skin later.

It's really very.

Wall.

Rich.

Well.

Yeah so we don't use eco flex because
we have found that materials have be

weak we can't manage to match the behavior
of eco flex to any standard material model

whereas other elastomer is do you it's
very odd you might have so in the plot

that I shared with stress relaxation and
we tested three different types of

elastomer is and two of the elastomer
show beautiful stress relaxation and

then one there's like one rogue line
that's just a flat line at the bottom

actually goes in the opposite direction if
you look at it closely that's eco flex.

We Ever read never been able to we've
looked at a lot we've never been able to

quite pinpoint that but
to get your actual question

I think one of I would say for this
particular robotic skins in particular

the idea is that there modular
kind of disposable in a way so if

it were to wear a break I would say just
get rid of it and slap another one on.

But I think more broadly for soft robotics
maybe and in the wearable application in

particular you know people ask why is it
washable you know that kind of stuff and

those are challenges that I think
the community is actively aware of and

working on.

All right.

Thank you.