How Robot Armies Will Save Our Future Cities
[Music]
imagine
being commissioned to dismantle a
radioactive
exhaust stack or to dispose of
millions of tons of water contaminated
with
radioactive material it is not a
hypothetical situation
today close to a decade after the 2011
fukushima nuclear disaster in japan
real people are facing these toxic
questions
have to handle danger safely so as not
to poison
a people a country or divorce water
supply
our world faces inhuman challenges every
day
whether industrial accidents natural
disasters
or self-inflicted catastrophes
circumstances that i call
inhuman because they compromise our
safety
and our survival our future world
needs in human help we will need
robots
until now engineers mathematicians
and roboticists have worked together
to build and move single robots like
this one
the model for this was simple at least
for me
when my son nikon took his first
steps two years ago i saw matt
a numerical sequence that played out
like computer code
called an algorithm isn’t that
what you see when you watch a baby walk
for the first time
turns out babies are the perfect
mathematical models to help us
understand the principles of locomotion
that underline
human agility but like my son
i was not content with baby steps
i wanted my research to take off running
too
a robot modeled after a boy is not
enough to do the work our future demands
we will need teams of robots or
collaborative robots
core robots and so we have studied
new models to understand how to build
more agile and collaborative robots one
model
to inspire their form and another to
exemplify their function
the first model was trotting around my
house with nikon
telly ever notice have your pet cat or
pet dog
has an almost uncanny ability to land on
its feet
or maneuver the obstacle course of
furniture and people
in your house turns out
four-legged animals exercise and
unparalleled dexterity
that enables them to traverse a globe of
value train
more than half of our world is
inaccessible
to humans and the real vehicles we have
invented
so to be most useful our robots
should take form as quadra pets
so
the second model was circling my house
in the dirt and armies social ants have
evolved effective competencies to
cooperate with each
other towards collective goals
collective phenomena
in social ants are the result of a
hundred million years of evolution
they work together to transport large
payloads that can be hundreds or even
thousands of times the individual ant
they work in homogeneous or
heterogeneous teams
with smart leaders to carry objects over
rough terrain to their nest
an impossible feat alone is made
possible
together so our quadrupedal robots
must function like ants with models in
place we could now ask a new question
one that still plagues us today how do
we mobilize
teams of collaborative leg robots put
another way
how do we get individual robots to work
together
to solve a problem when they have no
intuition
to react sensibly to changing
circumstances
we could just throw into one computer
all of the programmed instructions we
have created
for the single leg robots and then their
instructions overlap
hope they will cooperate doesn’t sound
good
does it here is why let me explain the
calculations necessary to
inform a single robot’s movement the
quadrupedal core robots we are
engineering
have a manipulator that looks like this
essentially an arm that enables them to
help humans
with manual labor intensive tasks
like construction manufacturing assembly
or disaster relief
each of these robots must decide where
it will move
how and how much force it will exert
on a target while maintaining its
balance
the execution of these variables depends
upon the control algorithms we create
for the robots remember
control algorithms are sequences of
computer implantable instructions
that address the decision-making problem
for individual
members of a team for example a set of
algorithms must compute
the required torque generated by each of
22 joint motors
during every millisecond of a single
robot’s movement from point a to point b
while accounting for any environmental
obstacles it could encounter okay
so that’s for one robot
now imagine a team of 50 collaborative
leg robots
working together to carry a container of
contaminated water
individual team members still have the
same variables
where to move how much force to exert on
their target
and how to balance plus an additional
variable
how best to coordinate its individual
movements
with others to play as part of a team
like
leader puller or lifter
that means we would need to design and
implement
control algorithms that compute required
torques for
50 times 22 joint motors
while accounting for thousands of
optimal decision variables
environmental circumstantial or
otherwise
every millisecond and if that is not
enough
all of these decision making algorithms
will have to be
executable in real time
okay we buckle down and do it right
like ants well we have a few
problems that ants have had about 100
million years
to solve at the root of them
is the fact that we cannot have a
centralized computer
to disseminate and coordinate the
control algorithms
like the ones i just described it
creates
a computational bottleneck too much
input
for too complex and output put another
way
one computer brain cannot interpret the
data from all joints
and variables and relay instructions
back
to robots for simultaneous optimal
function
we must instead establish a paradigm
shift from
centralized approaches to more versatile
approaches
that use distributed networks of
computational
algorithms to orchestrate sophisticated
core robot teams
each robot must have its own computer
and computational algorithm
that is resilient and versatile enough
to make
cooperative decisions based on its own
measurements and if each robot
function reactively it can efficiently
cooperate with
others
okay how do we do it the evolution of
robotic teams
that cooperatively manipulate objects
can be described by complex systems
complex systems are composed of many
components that
interact with each other in my lab
we develop what are called advanced
distributed control algorithms
based on optimization techniques that
allow
collaborative locomotion of leg robots
while carrying objects
but every step forward prompts
a new question and the search for
answers
leads me back to my yard where nikon
and telly are playing looking closer at
the ants i wonder
how can our mathematical algorithms
replicate
the biologically driven collective
phenomena
that these social insects have perfected
how can we advance the growing body of
research
suggesting that the animal locomotion
control is achieved through distributed
control schemes
how can we transcribe ant behavior into
control algorithms and why are those
instructions
into mechanics of core robot brains
answers would not only save lives
they would improve the quality of human
lives
in addition to disaster relief medical
applications
of distributed control algorithms
and collaborative locomotion include the
development
of robot guide dogs for the 1.3
billion people living with visual
impairment
and the development of powered
prosthetic legs
to stabilize accident victims and people
with disabilities our future cities
need robots teams of robots
to execute sensitive and sophisticated
solutions
to otherwise life-threatening
environments
to bring stasis and efficiency to our
ever
increasing human needs to collaborate
with humans to navigate an uncertain
future
that is guaranteed to include a
complexity only matched
by an interconnected network of
quadropodalco robots
how do we get there the answer
might be crawling in the dirt
you