[Applause]
according to the World Bank nearly
60,000 people die each year in natural
disasters mostly in developing countries
most of these deaths are due to building
collapses in earthquakes after an
earthquake areas can be inaccessible
making it difficult to search for
survivors and for survivors to find
potable water this makes robots and
drones the perfect tool to search for
survivors after an earthquake in fact
one of the largest drone manufacturers
in the world recently said that one
properly equipped drone can find a
missing person up to five times faster
than traditional search methods that’s
just one drone imagine what 210 or even
a swarm of 100 drones could do right now
each drone is operated by a single
trained pilot which tell the drone
exactly what to do imagine if instead
that pilot could control that entire
swarm of drones or even better if that
entire swarm of drones could work
autonomously on their own without a
trained pilot in developing countries
where a trained drone pilot might be
hours away local first response teams
can deploy an autonomous team of aerial
robots to search for survivors and tell
them where to go immediately after a
natural disaster saving many more lives
I’ve been working on these kinds of
complex problems of how to get teams of
robots to work together for more than a
decade and it’s deceptively difficult to
get robots to work together but we have
big dreams we want robots to search for
survivors after earthquakes we want
robots to contain forest fires we want
them to deliver medicine to area
in need but these environments are very
different than the environment that that
we’ve trained the robots in so we’ve
solved very similar problems to these so
in this example this is a video from my
lab where we’re doing a demonstration of
drones flying through windows so we’ve
been successful in labs in manufacturing
and warehousing facilities and even in
entertainment but these environments
like I said are very different and so
we’re dreaming big but something is
holding us back and I believe it’s how
we solve the problem it’s a really big
challenge to get a team of robots to
work together on search and rescue in
fact it’s the biggest challenge I faced
in my career I want you to think back to
the biggest challenge you’ve ever faced
did you solve it entirely on your own or
did you work in a group or ask other
people for advice I know that when I
work on a big challenge I ask as many
people as I can I try to get as many
opinions from people who’ve been through
it and people who have it people who are
my senior and people who are my junior
people in my field people outside my
field I collect as many ideas as I can
and then I take the best of those ideas
that work for me and so we work in
groups so that we can take advantage of
the diversity in our experiences and in
our knowledge so that we can come up
with better solutions but when I think
about how we actually try to get robots
to work together we’re not applying this
right now what we do is we have our
problem like search and rescue it’s very
complicated so we try to make it simpler
by making assumptions like the building
is made of a certain kind of material or
the earthquake is going to be of a
certain type or a certain size and then
we optimize the solution based on those
assumptions and we deploy it on all of
the robots the trouble is that the world
rarely meets our expectations
and so when one of those robots fails
all of them are gonna fail so we’re not
applying what comes a second nature to
ourselves trying to come up with diverse
solutions so that they work better and
so if humans are so good at this then
why can’t we learn from humans robot
assist hub is studying animals like
birds fish and sand lions to try to
figure out how they collaborate on tasks
so that we can observe them and actually
apply those learnings to robots but we
can actually communicate with humans but
unfortunately it’s not that simple as
you may already know robots are nothing
like humans humans have very different
ways of communicating just a facial
expression or a hand gesture can
communicate something to you robots
can’t do that their communication is
much less expressive and much more
intentional robots can communicate by
broadcast so for example if you heard
over the speaker’s right now somebody
saying follow me if you don’t know where
that’s coming from how would you know
what to do robots can also communicate
by address that’s like maybe one of your
friends in the audience sending you a
text message saying come over here now
as a human you might be able to figure
out where your friend is by using
context but robots don’t have that
capability and so communicating within a
large team of robots becomes really
difficult because they need to share a
lot more information than we’re used to
that uses up all of our bandwidth so
that we can’t use it for more important
things beyond communication
we’re must also see the world in an
entirely different way in this video you
can see the world from a robots
perspective it’s really hard for us as
humans to understand what’s going on in
this video but the robot is actually
using this information to complete its
task and so if we expect robots to be
able to use this small amount of
information to do complex things it’s
no wonder that we haven’t been able to
do things like use teams of robots for
search and rescue so I thought about
this for a very long time how can we
actually get humans how could he get
robots to act more like humans but we
can’t really do that because they’re so
different so instead what we decided to
do is to actually use interfaces to
change how humans act so humans can act
more like robots and so we develop a
multiplayer game in order to learn how
humans coordinate on complex tasks and
we gave them very robot light
capabilities so in this game each player
they’re all human players is represented
by a single dot on the screen there are
many other people playing at the same
time but much like you could only see a
limited field of view of people seated
around you we only allow them to see a
limited field of view and much like the
person next to you blocks what you see
beyond them we also include this kind of
occlusion we allow the people to move up
down left and right and then we ask them
to do tasks like form rectangles and
circles together note that I didn’t
mention that we give people any kind of
communication so we already talked about
why communication is difficult so we
don’t include any communication in this
game so to give you an idea of what it’s
like to play the game if you’re playing
the game you’re the red dot in the
center here and all the other black
parts are controlled by other people so
this is not a computer game this is
people actually playing the game and so
you can see in this small example that
it’s kind of you can kind of understand
what’s going on and so in a small group
with this little amount of information
it’s actually easy to form a rectangle
but now I’m going to show you what it
looks like in a larger group but to give
you a better understanding of what that
looks like I’m gonna show it to you from
it
overhead view so you can see all the
players know that the players here only
see that small circle around them as
well so we ask these players to also
form a rectangle and so in the beginning
what happens is because people can’t see
each other they start forming rectangles
in different locations and eventually
one guy in the large group in the center
figures out there aren’t enough people
there will try to go out and find two
other people but doesn’t go far enough
someone from the other group does the
same thing goes far enough I’m not as
able to actually go back and recruit the
other players using motion alone no
communication just motion and so people
played a game very differently so some
of them form groups some of them are
passive some of them are active some of
them are explorers and all these
different personalities combined to make
the game actually work and so from this
game we’re learning how humans
coordinate on these kinds of tasks so
that we can better apply this to robots
you might be asking me couldn’t I just
come up with a controller that would get
the robots to form a perfect rectangle
every time the answer is yes just like
if I could train a team of aerial robots
to search the same exact building that’s
collapsed after an earthquake and they’d
be perfect at it every time but change
that building change the location change
the materials it’s not gonna work
and this is why diversity is necessary
we need to take different solutions and
combine them so if we have robots that
are trained on different kinds of
buildings then one of them is bound to
work and the others can learn from that
one so instead of taking our problem and
making a set of assumptions and applying
it to all of our robots instead we
should come up with different solutions
that have different strengths combined
the best ones and use those on the team
of robots this way if one of them fails
they can actually learn from the others
and
this actually goes well beyond just
having robots form circles and
rectangles diversity is extremely
important and what we’re learning
through this game is how diversity works
by studying how humans use it and so
with this information we can actually
learn how much diversity is necessary
and whether it’s different for different
tasks so that we can take robots out of
our labs and warehouses and
manufacturing facilities and put them in
these complex environments like search
and rescue or any kind of outdoor
example so imagine in a developing
country after a natural disaster that a
local first response team can deploy a
team of aerial drones to search for
survivors and tell them where to go
those hours or minutes
even than it saves can help us change
this to statistics dramatically but this
goes well beyond search and rescue
having teams of robots that can take
advantage of diversity will allow us to
do deep-sea exploration they’ll allow us
to monitor our aging infrastructure and
even take teams of robots to space the
sky is the limit as the late Maya
Angelou said in diversity there’s beauty
and their strength our differences or
make us unique and will make us valuable
and we need to embrace that let’s teach
our robots to do the same thing so that
together we can change the statistics
thank you
according to the World Bank nearly
60,000 people die each year in natural
disasters mostly in developing countries
most of these deaths are due to building
collapses in earthquakes after an
earthquake areas can be inaccessible
making it difficult to search for
survivors and for survivors to find
potable water this makes robots and
drones the perfect tool to search for
survivors after an earthquake in fact
one of the largest drone manufacturers
in the world recently said that one
properly equipped drone can find a
missing person up to five times faster
than traditional search methods that’s
just one drone imagine what 210 or even
a swarm of 100 drones could do right now
each drone is operated by a single
trained pilot which tell the drone
exactly what to do imagine if instead
that pilot could control that entire
swarm of drones or even better if that
entire swarm of drones could work
autonomously on their own without a
trained pilot in developing countries
where a trained drone pilot might be
hours away local first response teams
can deploy an autonomous team of aerial
robots to search for survivors and tell
them where to go immediately after a
natural disaster saving many more lives
I’ve been working on these kinds of
complex problems of how to get teams of
robots to work together for more than a
decade and it’s deceptively difficult to
get robots to work together but we have
big dreams we want robots to search for
survivors after earthquakes we want
robots to contain forest fires we want
them to deliver medicine to area
in need but these environments are very
different than the environment that that
we’ve trained the robots in so we’ve
solved very similar problems to these so
in this example this is a video from my
lab where we’re doing a demonstration of
drones flying through windows so we’ve
been successful in labs in manufacturing
and warehousing facilities and even in
entertainment but these environments
like I said are very different and so
we’re dreaming big but something is
holding us back and I believe it’s how
we solve the problem it’s a really big
challenge to get a team of robots to
work together on search and rescue in
fact it’s the biggest challenge I faced
in my career I want you to think back to
the biggest challenge you’ve ever faced
did you solve it entirely on your own or
did you work in a group or ask other
people for advice I know that when I
work on a big challenge I ask as many
people as I can I try to get as many
opinions from people who’ve been through
it and people who have it people who are
my senior and people who are my junior
people in my field people outside my
field I collect as many ideas as I can
and then I take the best of those ideas
that work for me and so we work in
groups so that we can take advantage of
the diversity in our experiences and in
our knowledge so that we can come up
with better solutions but when I think
about how we actually try to get robots
to work together we’re not applying this
right now what we do is we have our
problem like search and rescue it’s very
complicated so we try to make it simpler
by making assumptions like the building
is made of a certain kind of material or
the earthquake is going to be of a
certain type or a certain size and then
we optimize the solution based on those
assumptions and we deploy it on all of
the robots the trouble is that the world
rarely meets our expectations
and so when one of those robots fails
all of them are gonna fail so we’re not
applying what comes a second nature to
ourselves trying to come up with diverse
solutions so that they work better and
so if humans are so good at this then
why can’t we learn from humans robot
assist hub is studying animals like
birds fish and sand lions to try to
figure out how they collaborate on tasks
so that we can observe them and actually
apply those learnings to robots but we
can actually communicate with humans but
unfortunately it’s not that simple as
you may already know robots are nothing
like humans humans have very different
ways of communicating just a facial
expression or a hand gesture can
communicate something to you robots
can’t do that their communication is
much less expressive and much more
intentional robots can communicate by
broadcast so for example if you heard
over the speaker’s right now somebody
saying follow me if you don’t know where
that’s coming from how would you know
what to do robots can also communicate
by address that’s like maybe one of your
friends in the audience sending you a
text message saying come over here now
as a human you might be able to figure
out where your friend is by using
context but robots don’t have that
capability and so communicating within a
large team of robots becomes really
difficult because they need to share a
lot more information than we’re used to
that uses up all of our bandwidth so
that we can’t use it for more important
things beyond communication
we’re must also see the world in an
entirely different way in this video you
can see the world from a robots
perspective it’s really hard for us as
humans to understand what’s going on in
this video but the robot is actually
using this information to complete its
task and so if we expect robots to be
able to use this small amount of
information to do complex things it’s
no wonder that we haven’t been able to
do things like use teams of robots for
search and rescue so I thought about
this for a very long time how can we
actually get humans how could he get
robots to act more like humans but we
can’t really do that because they’re so
different so instead what we decided to
do is to actually use interfaces to
change how humans act so humans can act
more like robots and so we develop a
multiplayer game in order to learn how
humans coordinate on complex tasks and
we gave them very robot light
capabilities so in this game each player
they’re all human players is represented
by a single dot on the screen there are
many other people playing at the same
time but much like you could only see a
limited field of view of people seated
around you we only allow them to see a
limited field of view and much like the
person next to you blocks what you see
beyond them we also include this kind of
occlusion we allow the people to move up
down left and right and then we ask them
to do tasks like form rectangles and
circles together note that I didn’t
mention that we give people any kind of
communication so we already talked about
why communication is difficult so we
don’t include any communication in this
game so to give you an idea of what it’s
like to play the game if you’re playing
the game you’re the red dot in the
center here and all the other black
parts are controlled by other people so
this is not a computer game this is
people actually playing the game and so
you can see in this small example that
it’s kind of you can kind of understand
what’s going on and so in a small group
with this little amount of information
it’s actually easy to form a rectangle
but now I’m going to show you what it
looks like in a larger group but to give
you a better understanding of what that
looks like I’m gonna show it to you from
it
overhead view so you can see all the
players know that the players here only
see that small circle around them as
well so we ask these players to also
form a rectangle and so in the beginning
what happens is because people can’t see
each other they start forming rectangles
in different locations and eventually
one guy in the large group in the center
figures out there aren’t enough people
there will try to go out and find two
other people but doesn’t go far enough
someone from the other group does the
same thing goes far enough I’m not as
able to actually go back and recruit the
other players using motion alone no
communication just motion and so people
played a game very differently so some
of them form groups some of them are
passive some of them are active some of
them are explorers and all these
different personalities combined to make
the game actually work and so from this
game we’re learning how humans
coordinate on these kinds of tasks so
that we can better apply this to robots
you might be asking me couldn’t I just
come up with a controller that would get
the robots to form a perfect rectangle
every time the answer is yes just like
if I could train a team of aerial robots
to search the same exact building that’s
collapsed after an earthquake and they’d
be perfect at it every time but change
that building change the location change
the materials it’s not gonna work
and this is why diversity is necessary
we need to take different solutions and
combine them so if we have robots that
are trained on different kinds of
buildings then one of them is bound to
work and the others can learn from that
one so instead of taking our problem and
making a set of assumptions and applying
it to all of our robots instead we
should come up with different solutions
that have different strengths combined
the best ones and use those on the team
of robots this way if one of them fails
they can actually learn from the others
and
this actually goes well beyond just
having robots form circles and
rectangles diversity is extremely
important and what we’re learning
through this game is how diversity works
by studying how humans use it and so
with this information we can actually
learn how much diversity is necessary
and whether it’s different for different
tasks so that we can take robots out of
our labs and warehouses and
manufacturing facilities and put them in
these complex environments like search
and rescue or any kind of outdoor
example so imagine in a developing
country after a natural disaster that a
local first response team can deploy a
team of aerial drones to search for
survivors and tell them where to go
those hours or minutes
even than it saves can help us change
this to statistics dramatically but this
goes well beyond search and rescue
having teams of robots that can take
advantage of diversity will allow us to
do deep-sea exploration they’ll allow us
to monitor our aging infrastructure and
even take teams of robots to space the
sky is the limit as the late Maya
Angelou said in diversity there’s beauty
and their strength our differences or
make us unique and will make us valuable
and we need to embrace that let’s teach
our robots to do the same thing so that
together we can change the statistics
thank you
Please follow and like us: