Towards the robots of science fiction
[Music]
[Applause]
robots
have long imagined a world where you
could
go out and explore the cosmos remotely
where you could have a system that would
help you restore mobility
restore function in systems that would
enable you to do things like monitor
wildfires and address them instantly
right
there’s a long history of our
imagination running wild with what
robots could do
right this is documented in science
fiction going back to the 50s and before
by greats like asthma then heinlein and
clark
and the idea was robots could be there
to be better versions of ourselves to
make our lives better
where are we today with robotic systems
well there’s been incredible
gains that have been made in the last
five or ten years we have
robots from companies like boston
dynamics that are running outside that
are doing backflips
this is really incredible this is
something that i couldn’t have imagined
20 years ago right and you’ve all seen
these videos probably really incredible
stuff
not only that but we have robots that
are going to start to explore the cosmos
we already have rovers on mars
we have a humanoid robot on the space
station going above our heads
and we’re going to be sending a
helicopter to mars in 2020
so we’re starting to take robots and
extend ourselves through them
finally we’re starting to see robots
helping people live better lives
so this is by rewalk it’s a lower body
exoskeleton for paraplegics
this is a paraplegic walking in the
device with the aid of some crutches
and again this is what we want to do we
want to use this power this technology
in a positive way to help people live
better lives but there’s one important
point here that all these videos don’t
show
so what’s the secret below at all well
first
in the context of this exoskeleton
there’s a hint
the hint is that it requires crutches
why does it require crutches it because
all the videos you just saw
however amazing they are were the one
time it worked
after a thousand times of trying to make
it work
so boston dynamics i give them a ton of
credit because they show
outtakes of when it didn’t work on their
robots
so it gets worse this is a bad day it’s
a bad robot day
so we also have the backflip which was
amazing and here’s some outtakes from
the backflip
again these are the thousands of runs
that you run before it finally works the
way it should
and just to make it clear i’m not
picking on everyone else but myself
here’s a robot at caltech cassie this
was built by agility robotics but we’ll
be running our own algorithms on it
and we we’re like let’s demo this so
some kids were outside of our building
and we put it in front of the kids and
they got really excited and they were
taking pictures and this is all
wonderful
and then just a face plant
so i’m not immune to this the fact that
it takes a lot of work to get robots to
work
[Laughter]
there’s a grad student running in i was
there i didn’t make it in it was my grad
student that came in to save the day
so so where does this put everything
what’s the point of this the point of
this is twofold first
we imagine robots doing all these
amazing things why aren’t they doing
them today
and i would argue the big thing that’s
happening is there’s a gap right now
there’s a gap between what we understand
mathematically about the world and about
robots
and what we can actually get robots to
do
and the ultimate goal of getting robots
into the real world is to bridge this
gap
to bridge this understanding so that
when we do something we can put it on
the robot it will work the first time
every time and at the core of making
that happen
is mathematics so let me get slightly
technical for one second
and explain this process of what robots
look like and why it matters well it
matters again because we can get it on
robots that walk outside
we can get it on robotic assisted
devices but we have to start
at the mathematical level so let me just
give you a microcosm we always see
robots doing these amazing things what
actually goes on here how do we actually
make these robots do things
well the first thing you have to do is
start with the equations you have to
start with the math
the physics describing the system so
what you see on the screen behind me
is a small set uh just a little
component of the equations describing
the robot in the picture these equations
are hundreds of megabytes long
they’re massive and they’re all
complicated you can’t really see it here
but they’re signs and cosines and
squares they’re non-linear
they’re really daunting in their
complexity so we have to start with
those though because those govern the
forward evolution of the system
once we take those equations we pair it
with what we think
walking should look like so models of
locomotion let’s say
and then what theorems do is they take
all that mathematics
and they distill it down to the essence
the understanding of what walking in
locomotion really
is and we mathematically prove that in
fact we can generate this walking on
robots
now that math is great of course how do
we get it to work in practice is the
real question
well what math really does is it gives
algorithms those algorithms
which you see running here in this
little screenshot generate the walking
behaviors that finally go
on the robot so that’s the process
that’s underlying all of these things
and let me just kind of show you an
example of this theorem in action
so this theorem is running on this robot
via algorithms that were developed by my
students and myself
at our research lab and and so what’s
happening here is this robot
is walking dynamically with everything
on board
in a natural human-like way now the key
point here is no human data went into
generating this walking in fact this
walking is just purely a function of the
dynamics that are inherent to the system
itself
in fact the walking comes from making
the robot walk in the way that we walk
that is having heel strike toe strike
and a toe lift in its feet
and then it also comes from the fact
that there’s springs in the ankles that
we store energy
and release and the result of the
physics interfacing with the mathematics
is this natural human-like locomotion so
what else do theorems give us what else
does mathematics give us
gives us generality so once we
understand how to do this on one robot
we don’t have to repeat the process all
over again
rather we can take a totally different
robot in this case cassie which i
mentioned
earlier and we can take and apply the
same procedure you don’t even have to
look at all the stuff in this because
i’ve explained it all to you already and
we can get walking on flat ground in the
lab
we can get walking even outside now in
this case on some flat ground
outside and then finally we can even get
walking on some rough terrain
meaning that this math and these
theorems are robust enough that they let
us do this
and in environments that we didn’t
actually plan for
in fact we can take really aggressive
terrain to a certain degree
so this is a root system at cal tech
right outside of our building
and you see that the robot’s kind of
dancing a little bit here it’s having
it’s giving it a go trying to walk
through this very tough terrain
until it falls it did pretty well though
right it almost made it
we’ll watch it one more time because
it’s actually really amazing and
remember all of this is the mathematics
in action we’re not planning here
this is just using the dynamics of the
system to try to walk
almost all the way so the other thing
about theorems that’s really nice is
it’s not restricted to only bipeds
in fact we can apply the same general
mathematics to quadrupeds
quadrupeds are basically just two bipeds
that are kind of working together to
carry this thing in the middle right
and so this is the same exact root
system that cassie walked on but now
turns out the quadruped can actually
handle it you’ll see it actually gets a
little bit of air time there in the
process
and then ends up finishing the finishing
and not falling in this case which is
very nice
so we can apply this to different
platforms which is really exciting
finally we can do different behaviors
this is jumping cassie again
in this case cassie jumps about seven
inches in the air meaning his feet are
seven inches off the ground
i think this is probably almost better
than i could do and again the point is
once we understand that this physics of
movement this this mathematical
understanding of movement
we can make robots and different types
of robots do amazing things
so the question now becomes why do we
care
this is really cool and these are fun
and the videos are great and it’s great
seeing robots do amazing things
but why do we care and i think the point
here is that with this
understanding with understanding comes
responsibility
and taking this to a a domain where we
can actually
help people live better lives and let’s
be real however cool robots jumping is
that’s not going to make your life
better right
well maybe for a second when you watch
the video but beyond that it won’t it
won’t help you all too much
but imagine if we can take and translate
this understanding
to making people walk better
and this is the this is the merit in
studying bipedalism
because now that we understand how to
get human-like walking on bipedal robots
at a fundamental and mathematical level
we could translate it to things like
prosthetics and exoskeletons
right because this is all math it’s all
general we can view the human in a
general way and have them interact with
the device
in particular we can actually do this
very concretely and very scientifically
we can start with let’s say a walking
robot and we can take this theorem and
translate it to the device as if it was
a prosthetic
and that lets us equally take the same
ideas
and put it on prosthetic devices so this
is an actual amputee
walking with our first generation
prosthetic that was custom built by my
left
we’ve since made these prosthetics even
better this is our one of our latest
generation prosthetic devices again it’s
kind of amazing that
my students have built this by hand they
machine this they put it all together
all the electronics all the algorithms
all through their hard work and then we
take the math the theorem and put it on
here
and again just like we could take robots
outside we could take these prosthetic
devices outside
so in this case two of my students are
walking with the prosthetic around
caltech
they can handle terrains they can handle
some different environments
so this is this is what we want to do is
translate these technologies
but we can take these ideas even a step
further and this is this is more
challenging
more fun so a few years back we started
working with this
startup company in france of of all
places
and they had this idea they bravely had
this idea that they wanted to take the
math that we developed for walking
robots
and put it on exoskeletons for
paraplegics
so they designed and built this this
hardware but we
took the math that we have for walking
and put it on the device itself so my
grad student is in this in the video
you’re about to see here
helping them test out some of these
algorithms now the important thing is if
we can make this walk dynamically on its
own
what can we do we can put people into it
and they can walk dynamically even if
they can’t walk themselves
so what you’ll see here is first the
demonstration that the exoskeleton can
walk on its own
with nobody in it so we have a dummy in
it and it’s walking dynamically
again through the same map but now we
put a paraplegic in it
so this user is a complete paraplegic
meaning he has absolutely no function of
his legs and has not for 10
years and this is the first example of
dynamic walking
crutch-free walking with a paraplegic
that’s ever been demonstrated
it’s a powerful moment when you can
realize that this this passion of yours
this mathematical understanding that you
obsess over all the time can actually go
out and help people you know i heard
stories of this clinical testing the
loved ones around them
really connected with seeing them for
the first time upright
walking in a natural and dynamic way so
this is where we want to take these
ideas
and in this context at caltech we’re
fortunate enough through this
collaboration that we have one of these
devices
at caltech now and so what we’re able to
do is test it
in action test new ideas in the process
and try to bring them to
a case where we can actually have robots
and people work together
so that’s the vision underlying all this
as we want an environment where robots
and people
can interact in positive and beneficial
ways
so it’s not really fair you got to see a
bunch of videos of robots walking and
don’t get to see one yourself
so i’d like to welcome cassie to the
stage
so this is this robot was built by
agility robotics but is running all of
the math that i showed you today
meaning the mathematics under the hood
is what was developed by my students and
myself and collaborators in our lab
and you see that it has a very dynamic
feel to its walking
that’s because again we’re leveraging
the dynamics of locomotion
so in this case you’ll notice that
cassie has very small feet
that’s why it has to keep moving it also
has springs so it has these compliance
that has to go through
to help balance the system so we can
even do things
so so we’re going to try first here and
head up on the
let’s go good job cassie so cassie got
under the
the tad dot this might be one of the
first robots that’s been walking on the
ted dot dynamically
and so it’s able to handle all of these
terrains again
exactly because of what i talked about
earlier taking these
concepts and generalizing them
mathematically so that we can
characterize locomotion and these are
the same algorithms we ultimately put on
exoskeletons
to help people walk better and you can
imagine
somebody who can’t walk now walking with
these algorithms and able to even handle
a little bit of disturbances
get pushed on a little bit so we can
start to get close to robots
again years ago i wouldn’t have imagined
being able to stand next to a dynamic
walking robot and actually put my hand
on it
and interact with it and that’s where
we’re at today and
that’s where we want to head tomorrow is
more and more of this
human and robots working together in a
positive way
to affect change so with that thank you
very much
[Music]
you