Congratulations, Your Robot Just Accepted Your Friend Request


Uploaded by GoogleTechTalks on Jul 27, 2010

Transcript:
>>
So, I'm Andy, and I'm sponsoring with Tech Talk by Rich LeGrand, Illah Nourbakhsh--did
I addressed it right, Illah? >> NOURBAKHSH: Yes, close enough.
>> Okay. All right. Illah's from CMU, and Rich is the head of the company called Charmed
Labs. You might have seen it. They've gotten together and they've built something that
I don't even want to attempt to describe but I'll let them do it in their talk. We have
about 40 minutes and then we should save a little time for questions. There's a dory
page that was on the invite, so feel free to add your questions to dory page. We'll
handle them when we're done, okay? Bring it in you guys.
>> NOURBAKHSH: So, thanks everybody for showing up, and thanks Andy for inviting us to come.
This is an honor to be here. We have a two part presentation. So I'm going to set you
up with the problem that we envisioned and have messed around with for many years. We've
actually been working together for more than a decade. And then, Rich is going to come
up and present to you the solution direction we've been going in and actually give you
kind of call to arms. So, it's two parts and it's about robotics and robotics' future,
and how we can make the most interesting robots possible. That's fundamentally what it's about.
And it'll be great if I could show you a, you know, some kind of Moore's Law graph for
robotics. Wouldn't that be cool? I can show that to you and say, "Here's where we are
and we can extrapolate from that, and this is how cool our robots are going to be, and
how many we're going to have in our lives in five years." But I can't do that. It's
completely impossible to do right now and in fact, there's good reasons why you actually
can't do it at all. And that's the subject of another talk about why that's essentially
impossible so we can't do that. We can look at the progress that has happened in robotics
to set some context stuff for what we're going to propose today and what we've been working
on. And if you think about the progress in robotics, there are some pretty big and hefty
robots that are pretty outstanding. I mean, they're pretty amazing in what they do. There
are also deceptively challenging issues that swirl all around them. So, I don't know how
many of you have read P.W. Singer's book "Wired for War", but the Predator drone is a really
interesting example. It's one of these robots that's--well, it's a robot in the sense that
there's very significant amounts of mixed control and mixed autonomy going on. So, there's
a human in the loop but it's got significant autonomy on board. It has its operation problem,
it's got distance problems, telemetry issues, and it doesn't touch our lives everyday. It's
not something we wake up with except if there's a news report about something bad happening
or something good happening depending on your political point of view. But it's an example
of something that matters a lot for robotics which is how humans and robots interact with
each other, and how we design interaction, and tell operation, and tell their presence.
So that's one interesting category of progress that you can point to. Another interesting
category of progress you can point to has to do with dynamics and control. This is a
big dog out of Marc Reibert's lab and company in Boston. It's pretty astounding because
one of the things that's really hard in robotics has to do with human muscular control. So
having series electric--series, elastic actuators, they can tune themselves, adapt themselves
the right way, and have the right energy then so it's actually can do things like walk and
jump. That's pretty much remarkable. And so, if you also think about robotics and Moore's
Law or some such thing for robotics, this is pretty important. It's also a little bit
scary looking. I don't want to actually run into that in the parking lot alone, deep at
night someday. But if you think about robotics in terms of the Turing test, if there was
such a thing as a equivalent of the Turing test from robotics, then it would be very
pleasing to see that we've actually gone to the point in robotics now where there are
robots that can perform at the capability of humans that are virtually indistinguishable
from the robots. So, I had to show this to you--it's just too cool. Oh, the hand came
down. So, yeah. So, dynamics in control, that's a big one too. The last one I'll give you--yeah,
they'll flailing away there just like robots. There's another part of robotics, the regime
robotics. That's really hard that there has been big progress on it. It has to do with
computation and sense of vision, right? Taking a whole lot of sensors and doing something
human scale but the human does it with two eyes and a brain. We put--I don't know how
many sensors on one car. It's kind of absurd but we're getting the fusion down, right?
We have computers that can deal with that, process it, model it, and make decisions about
how to do things like parallel park, so that's impressive. So, okay. We've got impressive
progress for very expensive robots. I just showed you three robots, all of which cost
many millions of dollars. What about the things we wake up with and live at home with us?
What about the Blade Runner like things running around at home, marching, and giving songs
to us while we boil eggs for our enemies who visit us, never mind. What about the things
at home? Are they interesting? Are they something that we care about? If you turn to home robotics
and look at progress there, we have things that are being massed produced. We have tooling.
People have forgotten how many manufactured robots they sell. Not in massive numbers but
they sell. But you know there's a problem with all of these, with Rovio, and Roomba,
and Aethon TUGs, and all such things. They don't pull up my heart strings, right? I'm
not excited and giddy about the next version of any of these that are coming out. I'm not
like really, really, curious to see in the next electronic show what crazy, cool, exciting
thing I'm going to order right away and stand the long line for in either of those categories
of thing. It cleans my house or it does just tell a presence or it delivers things for
me in my hospital. So, in a way, robotics doesn't have its soul. It doesn't have that
emotional center of objects that you would become totally attached to the way you do
with, well, your PDAs and devices, and your music players, and all those good things because
they weave themselves into your life. Another way of saying this is every time I see a venture
capitalists, they ask me this question. Every single one of them in the first 45 seconds
of conversation after they find out I'm a robotics professor, "What's the killer app?
What's going on?" They have a mentality in which they're trying to figure out the killer
apps of a company can make it, right? And there are reasons to think this is the wrong
question. And I'll give you an example of why maybe this might be the wrong question,
all right? In the examples, some of you may have seen or some of you may not have seen,
Bum Bot. Who's heard of this one? Very few people. Okay. Well, you need to know about
this. This is interesting. The great thing is all the examples I'm giving in this talk
are examples that I use in my robotics and ethics class so they're all ethically questionable
examples. It's pretty cool. Bum Bot was made by this bartender. He owns a bar in Atlanta,
Georgia and he had this problem. There are homeless people who sit on the sidewalk in
front of his bar. He didn't want them there. So, he took a grilling device for smoking
meat and put a high pressure water cannon in it and made it remote control. So now,
he can be inside his bar, drive this down the sidewalk, and shoot high pressure water
at the homeless who are sitting on the sidewalk in front of his bar. Now, the reason this
is interesting while completely ethically questionable or maybe not. Even though we're
questionable, we can be a little fine than that. It's because it's the power of what
happens when somebody becomes really inventive with today's modern equipment that you can
very easily get hold of at home, depot, and a couple of other places, right? You don't
even have to go to SparkFun to do this. Not even that far. And so, my proposition--and
I think Rich's thesis with me is going to be, maybe what is the killer app is completely
and utterly the wrong question and maybe there's a totally different question. This is the
right question which is a rhetorical question, "What is your killer app?" And if people can
be incredibly creative like that with relatively low competence parts, building a robot does
not really that sophisticated. The interesting question is what can we do if we could reveal
to people a set of tools that allow them to make really, really interesting robots and
how far could they take it? So, the question I'm going to answer in the next three minutes
is what's the right recipe? What do you need to make magazine sophisticated robots, interesting
robots, robots that you would have an emotional attachment to in your home or in your social
sphere at the human scale? So, here are five or six things, really fast. One of them is
an ecosystem. Now, there's a picture of honeybees here because I keep honeybees. I keep a bunch
of hives. And it's an excellent example of a community in which there's a fantastic ecosystem.
If I want to buy the parts, there's a mix-and-match set of parts available for multiple vendors
that allow me to build up my hive just the right way. And there's a massive amount of
hardware and know how available on the Internet. I can order it, get it, and build these things.
So, there's a kind of physical ecosystem that makes it possible for me to put together sophisticated
beehives very quickly. And that kind of ecosystem is something we clearly need so that people
have the right building blocks. There's something else about building blocks though. We want
configurability, right? People put things together but we don't want tyranny. I run
a ton of educational programs around the world. We interviewed girls in middle school about
the LEGO robot programs and they don't like it. And you ask them why and they said, "Oh,
that's easy. LEGOs are ugly." They don't like LEGOs. Now, granted there are mostly men in
this audience so, you know, maybe the women in the audience can let the men know if this
is true or not. But there's an aesthetic components here. You need people to be able to use the
materials they want to use and fabricate something that's actually interesting to them. So, you
can't tyrannically suppress form for the sake of easier configurability and yet you want
configurability, so there's a really interesting challenge there that you have to solve. Another
one that I think a lot of the commercial robots are missing is emotional connection. The thing
the robot does and what it represents to you has to be in your heart, right? It has to
be interesting to you. This is a great example. This six called Braille tutor. It was part
of Bernadine Dias' program called TechBridgeWorld at CMU. And what they did which was genius
is the way you teach a blind child to write Braille is they use a device that pushes down
to make bumps. But your pushing down, so the bumps are on the on the bottom, right? So
you push down and you make the bumps using a stylus then you turn it over and you read
it. So, you have to learn to push with the stylus backwards, right? You have to write
in reverse. And so, you have to learn to write, learn the letters, and learn it all in reverse
first then split it around. So they made a robot that exactly takes the reversing part
out of the equation. So you write with the stylus and you can see physically in real
time what you've done. Just a very simple robot. Interesting in terms of emotional connection
because what they've taken is very simple parts that you can buy at SparkFun and a few
circuit boards you've fabricated, and they've made something that they feel really strongly
about. It has an impact, they smile about it, they feel good, they feel good in the
developing world as result too. So, you need the emotional connection. A vibrant community,
another one. We're almost done by the way with the list. Soon, it's Rich's turn. This
is a big deal; you know about that quite from the point of view of your Android and from
the point of view of every kind of form of social networking, and sharing that goes on
in the world. It's a little bit missing in the world of hardware sometimes. The example,
I put a picture up on here is we have an electric car program called charged car in which we
train mechanics and independent garages to convert cars to electric cars. And it's hilarious
because we've had events where we show who killed the electric car, we show gas land,
and we invite people to come from the public, and we bought an old Exxon station. So, we
run the whole program in the Exxon station which is kind of cool because we have an open
flame, we have a barbecue. You don't usually have open flames in Exxon stations. You'll
understand why. You're Googlers. Anyway, what's interesting about that is you get environmentalist,
activist, complete gear heads who couldn't care about the environment all in one room,
passionate about a particular subject. In this case, about electric cars and cars. Last
weekend, we had a formula F1 lightning, electric car in there. We had five different electric
cars in the room at the same time and people are all just all over them, talking about
planning and striking up conversations. So, fundamentally, you need a community but you
want to get diversity. You want people from completely different backgrounds to end up
working together and hacking together. And that works really well for a topic like cars
that people get. Unfortunately, that's another male dominated one when we do these events
but you need to do that around robotics in our opinion. Last, inherent power. This is
where a lot of the kids on the market fall down. They don't fundamentally give you enough
power to do really interesting things. This is just an example from Bug Labs where, you
know, if you're going to do some networking, how cool is it to have somebody that's running
full blown Linux. So, you use a real processor. You can actually code it up. It has its own
operating system. And it's often easy for people to get trapped in the world of microprocessors
and get far enough away from true competence that they can't do really sophisticated interesting
robots with vision, speech, and speech synthesis, and all those such. So in a nutshell, that
sort of--the set of things that we think you--if you could nail all of those at the same time,
then maybe we can answer the question, "What's your killer app?" We have an existence proof
that Rich wanted me to share. It's a good existence proof because it's exactly what
Rich and I did last. So, the thing we started working on before this was a project called
GigaPan. And I don't know how many of you are aware of GigaPan or what it is, so like
5% of you. It's basically, a lot of software and some hardware, and it's a robot that holds
anything from a DSLR to a small camera, robotically moves it around, takes thousands of pictures
with it, and then there's a stitching software that puts that all together, fabricates it
into a giant panorama, and then the website that takes billions of pixels and shows it
to you, so you can see Dubai in 45 billion pixels. Forty-five billion pixels is interesting
because you can zoom in, right? You get lesser detail. And there are obvious things you can
do with it. But what's interesting is to show how this works. When you give people enough
power and you have a real community, what do they do with it? So, I'm going to tell
you about three scientists really fast and what they did with this to show you diversity,
and I think this can happen on robotics. So, these are scientists, right? These were not
gear heads. They took the GigaPan and they did specific things with it for their science.
So look at how different what they did to this. Jane Goodall, they took tobacco plantations
in Tanzania. They take pictures of them because there are illegal plantations where the primates
have been pushed--shelved off because the trees are dead, right? They show the panorama
to their funders who don't have the time to travel to Tanzania. They showed this to the
funder and then they zoom in, and zoom in, and zoom in until in the corner, in the forest,
they can show every tree being ringed to be killed so they can increase the plantation
size and then they ask for money. So this is actually money requesting process, right,
using technology. Example number two, Alex Smith, University of Guelph in Canada, DNA
Barcoding of Life. He runs the entire Barcoding of Life project. They go around the world
and DNA Barcode species of insects, and capture with a GigaPan the entire context of the insect.
What's cool about it is they're storing the DNA now together with a link to exactly where,
say, that spider was in the entire forest. And then you can go and do a biodiversity
assay, counting every insect in the picture which is crazy. It's mind blowing. It changes
the way they think about biodiversity and lab work. Last example, this is where the
gear has to go and take over in kind of a hilarious way. So, when you do a scanning
electron microscope image of a part of an ant, it's great. It's really beautiful but
you just get one part of the ant. What if you could do the entire ant? Well, that would
be thousands of pictures, right? Who can sit there with an XY station and move their hand
like that for 50 days? So, what they did is they took the GigaPan, right? That same unit
I showed you. They took the motors in it, stuck rubber tubes on the axels and the motors,
pull the XY levers off the axels, stuck the rubber tubes in the axels, then they literally
program the GigaPan; top left, bottom right. It go. That's all. The GigaPan spun the XY
stages of the scan electron microscope and took a multi-billion picture of an entire
ant. So now, you can zoom in on any part of the ant and get the individual picture that
you would ordinarily be seeing. Crazy for nature photography and science engagement
because you're getting the big picture and you're getting all the small pictures you
could possibly want and may not even know that you want at the time when you do the
imaging. So, that's just an example of what happens when you take creative folks, scientists
in this case, and give them a technology that's engaging with enough power to do something
interesting with it. Okay, back to reality. So, what I do is for robotics, right? We want
everybody to make their killer app in robotics. So, how do we nail all these? Our proposition
is to nail this. First of all, to do ecosystem and configurability, you have to have a part,
right? Somebody who's going to mass produce parts for you. And we've been banging on this
for about two years now carefully. We lend them a partner, signed the full licensing
deal with them, and Rich will tell them about you. There's actually the biggest robot manufacturer
around so it's very exciting. So, we've got that done. We have an ecosystem and we have
configurability, and that's a really big first step. Next step, you need inherent power.
You want a processor that's absurdly good. And Rich will tell you about the one we have
now that's out and the one that's coming out that's even better. And last thing I'll say,
how do you nail this? How do you get an emotional connection and get a really good community?
Well, one possibility is stop thinking about the device that you're making as a device
and start thinking of it as something that's thickly integrated with you socially in your
environment, in your cloud. So maybe the problem is the robot and the you, and your world have
to be integrated together socially and technically in terms of data. And so, that you can use
essentially the computational cloud as a computational orthotic that helps your robot be far smarter
than it could be on its own. And so that you and the robot all feel like you're friends
with each other in a really deep way. So that's where I'm going leave it and the challenge
for Rich now is going to be sharing with you where we've gone. Now, I'm going to tell you
a few words about Rich before he goes up here because I respect him more than a lot of people.
That was really poorly said. What's a better way to say that? I've known Rich for a very
long time and I respect him deeply. Rich is one of these guys who's an inventor and he
knows how to put his finger on the pulse of the most emerging technologies. And before
they're even quite out, convince the company to give it to him so he can use them and invent
a completely new way of putting the pieces together that hasn't been thought of yet.
And so it's exciting because from a hardware point of view and the firmware point of view
he's been able to nail how we can redefine leading edge of robotics for people like us
to invent with. So anyway, having said that I'll stop now, and Rich, would you like to
tell us about the solution side. >> LEGRAND: Good stuff. I've never done any
talk with Illah before. We never rehearse this talk before and I'm not--I'm a nerdy
guy, right? I guess Illah is pretty nerdy too but I'm not so good at the presentation
stuff, so bear with me. I want to talk a little bit about the Telepresence Robot Kit which
was a project we did back in '07. It was a project between Charmed Labs and Carnegie
Mellon. It tries to take advantage of every person's you know inventive power, kind of
what Illah was just talking about. And essentially use Telepresence as the hook. Try to get kids
interested in Science and Technology. We got some money from Google, Intel, because they're
kind of scared about the pipeline problem, you know, there are not as many Computer Science
graduates nowadays. So, people invented stuff like, you know, they want to check on their
dog, they want to conduct their own, you know, there's a picture of a Mars Rover. They want
to conduct their own Mars mission in their backyard, you know, see what kind of life
forms are. They want to invent a robot that, you know, scares the heck out of their brother,
you know, that kind of stuff. And Terk has been very successful but we want to expand
it. So when we first got started, we had to create our own network and we got pretty bugged
down in a networking problem because we had to create our own custom peer-to-peer technology.
So what we used were relay servers. Essentially, you know, if your device is behind the firewall
or a NAT router, there's no way to talk to it but with the relay server, you can connect,
you know basically any two entities in the internet but that had scaling and bandwidth
issues, it had latency issues as well, and it's not really where we wanted to spend our
time. But we had to do it because there's really no solution that exist that allows
you to communicate with devices over the web. So, essentially the web is broken for devices,
there's plenty of middleware solutions out there that allow you to, you know, it's like
PubSub kind of stuff but they kind of assume that you have a fully connected network and
the internet isn't a fully connected network. There are existing push technologies out there
that might allow you to talk to a device behind a firewall but they tend to be unwieldy and
devices have, you know, some special needs. So, we kind of invented this term "Web embedded
devices." And essentially what it is, is we're assuming that these devices are you know pretty
small, 32 bit processor, you know, a couple of megabytes of memory. The targets must be
accessible regardless of the topology of the network, you know, they could be behind firewalls,
net servers, that kind of stuff. Good bandwidth, flow latency that kind of--we want to avoid
using really servers if at all possible. And we don't want to do any like crazy network
configuration stuff, we don't want to, you know, have to, you know, do any port forwarding
or assume that, you know, universal plug and play is enabled or any of that crap. It should
just work. So we sort of stumbled on this. So Gtalk and libjingle. We needed a technology
that, you know, dealt with NAT routers and firewalls and Gtalk does that. And since it
uses XMPP, we can, you know, push RPC, you know, Remote Procedures Calls through it and,
you know, remote procedure call is kind of the work force of embedded communication and
control. It also allows us to do a direct UDP connection often times between two entities.
They'll do like whole punching schemes that kind of stuff and get us a direct connection
and it does all that kind of transparently. So it's--and in the case if that doesn't work
they'll actually, you know, insert a message relay. So it's sort of almost there. So, I
wanted to quickly illustrate an example because I'm kind of bending some used cases here.
Imagine you have a small embedded system, it could be a Linux single board computer,
it could be an Android platform, it could be--your PC running windows or a Linux. Let's
assume that it can affect the environment some way or sense it because you wouldn't
want to embed it otherwise. The installer software it, you know, it runs upon boot up,
you know, it's just a little server. We make no assumptions about the networking only that
it has web access. It could be any kind of--any kind of web access and then it's up. Essentially,
you know, we're kind of almost there, you can go to your Gtalk client and type in some
thing you know, like get power status or list modules or send picture and then you'll get
the information. Well--but there's a lot that I'm not being, that's not being said about
implementation. You want to have, you know, again, RPC implemented on both ends, both
from the client and the server side. You can interact--interacting from a Gtalk client
is great. We want to be able to repli--mirror the RPC calls and text as well, so that you
can interact through Gtalk client as well a web browser. And we also want client libraries
that make a writing client apps easier because, you know, Gtalk clients are great and web
browsers are great and ubiquitous but we want to have, you know, be able to control these
things from custom applications. And we want to be able to do all this easily, we want
to able to set it up and configure it easily. So how could you use this? I mean, how useful
is this? Sensor networks; sensor networks' kind of a hot topic, it would be very easy
to create a sensor network with this technology. Environmental sensing is kind of a hot topic.
Social proxies; a social proxy is a device that gives you some kind of social interaction.
And, so for example a simple social proxy would be like a ring that you wear and it
glows when your girlfriend or boyfriend are rubbing their ring, you know, it's just like
a simple social interaction. A more complicated stuff I'll be talking about later. Smart home,
you could use this thing to you know, monitor your alarm, your thermostat, you guys have
heard all those scenarios, unlock, you know, your front door for a guest or something like
that. Robot telepresence of course; this idea of embodied interaction. You're there, and
able to manipulate things in the environment. Or your pet project, you know, essentially
this allows you to embed any kind of, you know, computing and talk to it anywhere. And
so, this is my pet project. This is a really killer app by the way. So Lance, he lives
in Austin, he's known to ride certain routes in Austin. So, what we have is a device disguised
as trash that sits on the side of the road and it takes pictures of bikers, you know,
you could use like some kind cloud based computing to recognize bikers, and faces and that kind
of stuff. It forwards the pictures to you and you look at and go, "Oh, that's not Lance,
that's not Lance." And then when you see him, you suit up and then you intercept him, right?
At some rendezvous location, and then you enjoy a ride with Lance for as long as you
can keep up with him. So, anyway, this would be really cool. I mean, some of my friends--so
I bike in Austin and a lot of my friends have biked with Lance. It's kind of frustrating,
they just kind of run into him. And so anyway, I'm going to show them. I mean I'm going to--he's
probably going to kick my ass but he's going to see a lot of me. Here's another idea, this
is an idea of a social proxy idea. Essentially, you have a device with a speaker, a camera,
network connection, a treat dispenser, a little robot treat dispenser, actually there's a
video that goes long as this. Oops... I don't know, has anyone seen this video? This is
funny stuff. So essentially this guy built this ball launcher for his dog and he can
sit there and play fetch with him all day, the dog gets it. So this device also has one
of these ball launchers on it. And so what you do is using your cell phone you call your
dog, you dispense a treat, you know, in a little button, you know, you go, "Hey Fido,"
and you dispense a treat. And then with a throwing motion, you like throw the ball with
your dog right, it has a little accelerometer in your phone and you sit there and play telefetch
with your dog while waiting in a stoplight. So...
>> [INDISTINCT]. >> LEGRAND: And the dog what?
>> [INDISTINCT]. >> LEGRAND: You just--yeah, in this case he
really loves the machine but what's interesting about this is, essentially you're putting
your dog on the web. I mean this device is acting as a proxy for your dog, a social proxy
and allows you to interact with your dog. It allows other people to interact with your
dog. You could have, you could put your dog on a social network for example and then you
can have, your dog could have his own friends that you approved, that you approved of course
and... >> I have a question.
>> LEGRAND: Yes. >> This would work with the robotic dog.
>> LEGRAND: It would work with the robotic dog, yeah. That's a good point actually. And
what else can I say? It's a--all right there is some other point about this. Well anyway
the, I'll go on to the next example. Another example would be a jelly bean dispenser. This
is another really good killer app. Essentially what you do is you where your social network
friend gives you a jelly bean. It could be a combo like, you know, two roasted garlic,
one Tabasco, one jalapeno. Your response, you know, "What's this, chilly?" And then
your--yes, he response, "Nope, this is hot wings." And so, this is like another good
killer app. So, you should have seen the ones that we didn't present, so. This is a good
app actually, I mean we've--TelePooch.com is not taken, so we're looking into it, maybe
investors but, no I'm just kidding. Please. Another--we have other ideas, Rover-like robot
with a camera and grasper essentially, you conduct your own Mars mission like I was saying
before, find and identifying animals, insects in your backyard, take pictures, you know,
send it to a blog automatically. A scavenger hunt, search for and collect items. Escape,
you know, you could rig up some kind of elaborate scheme where you press the button and it opens
the door where you get a key, you know, you have to take the key and then open the open
the door and then you escape. Not to fine China; devise a way to break you know a nice
fine China plate which actually was acted out in U.S. kitchen. A guy from Spain it actually
gotten on his table and knocked over a plate but luckily it also knocked over the robot,
so it rendered it incapacitated. So I'd like to talk about a little about the hardware
and software ecosystem that we've put together. So innovation first is a--they're well known
for their VEX line of robot products. They're also well known for being the hardware providers
for U.S. First which was started by Dean Kamen of Segway Fame. They actually ship more robots
than I-robot. They have some robots, they have some consumer grade robots at Target
and Toys R' US. And they also--the VEX robots are kind of nominally geared for education.
We've partnered with them. They provide the manufacturing and distribution of the controller
as well as lots of other products. >> [INDISTINCT].
>> LEGRAND: For a minute, I'm going to hang this around. That's the--that's the controller
that innovation first, Charmed Labs and Carnegie Mellon has designed. We've put together, you
know, kind of a generic operating system called TerkOS. It's based on OpenEmbedded. He has--uses
a Webserver for configuring network encryption. Your password, the name of the robot, that
kind of thing, it has drivers for motor control, general purpose I/O, interrupts, sensor input,
you know, all audio voice senses, as well as cameras. It also has a library for motor
and sensor abstraction. We have our own intergraded development environment which is based on
Eclipse. Uses, you know, it has a really simplified workflow for a high school audience, you can
click on, you know, connect to a certain--it has buttons on the right hand side, or left
hand side that will allow you to, you know, connect to a certain robot. You know, build,
debug, you know, Eclipse is kind of unwieldy so we simplified it. It also installs on Windows
and Linux. So we also, some of the work--so that's some of the work that we've done, some
of the future work that we need to do. We need a piece of software that runs on a device
or robot. Something that's pretty light weight. An Apache-like server that can handle RPC
calls, deal with video and audio streams, a text interface for debug and config that
mirrors the remote procedure call interface like I was walking about, some kind of loadable
module model, and it has be, you know, really easy to configure and set up. As well as client
libraries, we need client libraries to create an application, you know, create applications
that talk to or device/robot. Clients, you know, could run on PCs or possibly Android
platforms. We also--so these client libraries will manage the connections, dispense the
RPC request, decode and render the audio and video. At first we're going to add, you know,
C++ and Java support and then add, eventually down the road, JavaScript so we can try to
embed as much functionality in a browser as possible. We want to create lots of interesting
robot examples that uses technology, you know, possibly use Android phones to control the
devices as well as use the Android phones as targets themselves and we're just kind
of getting started. We haven't--these are just kind of the first steps. Here are the
specs on the Terk brain, the VEXPro is being passed around. It has an ARM9, 200 Megahertz
which is kind of passé nowadays, but that's why we're designing a new one. Has a 32 Megabytes
of DRAM, 16 of flash which, you know, that gives you just enough to have some pretty
interesting Linux functionality. Two USB host ports, WiFi, webcam support, it has these
motor controllers that use the back EMS signals from the motors for a feedback. So essentially,
a lot of motors need like optical encoders. These use the back-EMF voltage to actually
sense the voltage and they actually can integrate the voltage to give you position. And we're
able to do that because there's a field--an FPGA self programmable gate away--gate array
on, on board that implements these functions. We've essentially, you know, created a robot
ASIC with a bunch of robot IP on it and it's all open source you can go look at it. The
nice things is, is that this robot ASIC we can kind of targeted for different platforms
and make it easy to, you know, go to the different processors and that kind of thing. It has
a built-in speaker, has a keypad, and LCD, it's compatible with all the Vex motors, sensors,
beams, and it has--it's pretty well cost optimizes, it's pretty cheap. So, in the future we're
working on a Cortex A8, which is, you know, a lot of Android phones obviously use this.
Has a lot more RAM, more Flash, what's interesting is that it will have two camera inputs which
means, you know, we'll be able to, you know, these are regular camera inputs we'll be able
to--with all this process and apparel we'll be able to do a lot of real time image process
and provide a much more rich set of applications that you can handle. We'll also have an interesting
code branding deal with Texas Instruments, Texas Instruments is the source of the processor.
They're given us a lot of parts including the processor basically a CoS, we're going
to be able to provide this at a ridiculously low cost which is very exciting and of course,
you know, it will have enough power to run something like Android. So I want to go a
little bit into, you know, the ecosystem. These are the Vex parts that the Innovation
First guys manufacture. Here's the picture of all the gears; you have spur gears, pinions,
it has sprockets, chains, you have tractor treads, you have racks, you have rack gears,
you have worm gear, as well as double gears, lots of different wheels, you have regular
wheels, you have, you know, these, these universal wheels, omnidirectional wheel, these things
are called leg-wheels. There's a lot of different structural members, plates, axles, fasteners,
just all sorts of stuff, kind of reminiscent of the erector sets. But the nice thing is
you can both step on to it and you can bend it, so it's kind of nicely inherently configurable.
There's a lot of sensors there's encoders, light sensors, sonar sensors, exelerometers
bunch of different types of motors of different torques and sizes. Tactile sensors, reflective
sensors, controllers from eight bit to 32 bit, they have remote controllers. That controllers
which is like a, you know, PSP controller is actually has a Wi-Fi dongle in it. There
is a Wi-Fi dongle that it uses--that Wi-Fi dongle also use on the VEXpro. There's battery
chargers, there's batteries of course and then these are pretty cool, these little pneumatic
valves that are electronically controllable. So you can make a robot that, you know, can
punch or throw or kick or something like that which is could be kind of dangerous so they
have safety goggles. And here's example of various robot that you make. This is--both
pulled from the website. A lot of these robots were made by kids and here's one that shoots
rubber bands, has tank treads. Along the same thing we have a robot that shoots a paint
ball gun. I don't know if the kid actually made that but you can see it has like one
of those x10 cameras on there. This one, I've no idea what it does. Something, it had to
do with beer and pouring it, but I don't think a kid invented this one. I guess at this point
I'm going to turn it over to Illah because Illah you have--we have the deal, this is
the exciting stuff. >> NOURBARKHSH: So this is the part where
we're going to finished and the question was, how do we recruit you, right? And what is
it we are recruiting you for? We feel like we got enough of the basic ingredients that
is interesting now for an explosion of activity right? Thing number one is of course we'd
like to get some full time people working and developing some of this stuff and that's
called funding and that's not a conversation for this crowd. Thing number two is, you can
imagine short events. We can imagine doing robot camp where we do one out here, we do
one in Pittsburg invite for people to come up for two days and give them all the parts
in the world and let them go crazy doing a Hackathon. But we wanted to give you something
really visceral here now today, so that's thing number three. I grew up in Kansas City,
Putnam and they had these things called smorgasbord, which were absurd events involving ridiculous
amounts of food, but it was all you can eat and the food was pretty scary looking but
the all you can eat part was really interesting. So we come up with this idea for you. It's
an all you can eat deal for you so here's the idea, all right. Come up with what you
want to do with these robots and we'll select some of you and of course if you get lots
of good ideas we'll select lots of you, actually. And we're going to giver you boxes of parts
we actually came with two boxes full of VEX parts here, also, its building materials are
in these and there's the actual VEXpro controllers, so that's kind of cool, righ?. We'll give
you a box of things that you can get off the internet and things that you can't actually
buy yet because they are not for sale yet for three-four weeks but we'll just do that.
So, if you have an interesting idea we'll give you stuff so you can build it and the
dream of course is you going to do something with Android in the robot. That's the whole
interesting part but this is the best part which is we're going to give you a coupon.
It's an all you can eat coupon that you can use on the internet to get whatever you need
from Innovation First for free. So that's, that's got cache, right? Nobody in the world
has those yet. They have just been invented by Rich and Innovation First. So, that's the
proposition to you is we can give you all you can eat coupons. I think we're ready to
just take questions and see what you think of this general direction. Thank you very
much. >> So, let's check the dowry page if we can,
or we're going to start with a questions from the audience then we'll figure out how to
get the dory page projected. >> So this--the brain part it looks pretty
interesting. [INDISTINCT] the brains where here, the one thing I can't do with this is
it [INDISTINCT] around, it doesn't go anywhere. Has anybody--have you guys thought that maybe
just having a simple plug the phone in and then you can access to gears and motors and
structures? >> LEGRAND: Are you Ryan?
>> I'm Ron. >> LEGRAND: You're Ron?
>> I'm Ryan. >> LEGRAND: You're Ryan. Okay, okay. Yes,
yes, okay. I'm glad you made it. >> The boxer [INDISTINCT]
>> LEGRAND: Okay, cool. >> I just sat in the chair. I don't know who
these guys are. >> LEGRAND: Well you sit next to the good
dude who knows. What I mean so we actually, I mean, I agreed that's a very good platform,
I mean we're not involved yet as far as, you know, using the, you know, an Android handset
as a robot controller. But that's like a pretty sexy platform as Ryan will explain.
>> That's should be good. . >> NOURBARKHSH: We're good at introducing
each other. That's--one other part to that puzzle though. [INDISTINCT] for the studio
audience you take a picture of them. There's one other part that puzzle though which is
you want something that knows how to do motor control. The FPG is very useful for that.
You want something that can deal with the messy part of physical interaction of sensors
and motors and so the dream is makes sure we can optimize something, cost optimize something
for that. Whether the Android sits on the robot or whether the Android is a social communication
media for talking to it is another question it's interesting.
>> There's another there's another group of [INDISTINCT] which is the controllers of Android
can sit on the back end. We have this thing, [INDISTINCT] developed by [INDISTINCT] device,
when we talk about device message [INDISTINCT] device messenger, right? It uses lot of same
technology that you guys were talking about it's just that you're talking in general.
So it becomes common network and part of your social app, it just interface with your app
and then reach out to these guys through the existing network that the Android [INDISTINCT]
use to [INDISTINCT] and so maybe afterwards we can discuss and talk about the opportunity
[INDISTINCT]. >> LEGRAND: That's one of the [INDISTINCT.
I think the only, the only bias we had at hardware is cost, I mean Android handsets
for what you get is an incredible cost and you know, you get a lot, so that's great.
We want to get--our dream is actually get something down to like, you know, 99.95 or
something like that and we can do that just because, you know, because, you know consumer
electronics bring everyone's prices down. >> Instead of parts that you give [INDISTINCT]
there's a certain, certain assumption that we have to use [INDISTINCT], screwdriver [INDISTINCT]
really small kids [INDISTINCT] or safety issues [INDISTINCT] not sort of a age range you have
in mid minds the [INDISTINCT]. >> LEGRAND I think with VEX, you know, it's
probably, they target like, you know, 12 and up. And our stuff, you know, like the VEXpro
controller, you know, is even it's actually targeted for high schoolers. But, you know,
I mean, that's just, that's just kind of where we are right now. We want--once as the software
gets more and more sophisticated--the interesting thing as the software gets more sophisticated
we can bring it to more people. Stuff like the App Inventor, you know, is would be great,
you know, we can actually, you know, we could open up to big much bigger audience.
>> NOURBARKHSH: We actually had a program called robot diaries for middle school girls.
And there, the construction material is, is actually craft materials because there are
comfortable with it and they can do aesthetically really good robots with it. So, we're giving
them everything from feathers, to bells, to pipe cleaners and hot glue and they make amazing
robots and we have a processor, and we use--we had [INDISTINCT] programming language, but
now there's App Inventor. So you can totally take this in the middle school even and really
starts to hit some under [INDISTINCT] audiences in this domain.
>> Do you have idea [INDISTINCT] projects? >> NOURBARKHSH: We've done lessons plans for
robot diaries for three-four hour lessons and for full on three week long lessons and
we've deploy them. Actually the initial [INDISTINCT] just give us a funding to do it across the
public school system in Pittsburgh. So, we're going to be piloting across the whole public
school which is exciting because you have to learn how to talk teachers talk and had
it align to their curricular standards. So, that we're doing for the next three years.
>> So for robot diaries you're taking class advisory, you're teaching kids how to make
robots out of there--is there any--[INDISTINCT] actually getting kids together that have those
supplies so that people won't have to be [INDISTINCT] online instruction without defining [INDISTINCT]
can we just get them to make their pretty robots, some things that they like.
>> NOURBARKHSH: Yes the eco-system problem, you know. If VEX provides these stuff, I'd
love to see scholastics provide the craft box that you need to do robot diaries for
middle school. And I think that's totally doable, ones you have a--ones you can demonstrate
a market, they'll only be too happy to sale the academic supplies to schools.
>> There's that original company that sales like the [INDISTINCT] Oriental Trading Company.
>> NOURBARKHSH: Oriental Trading, yes they sell one of the original supplies. So it is
like getting a unique scope from them for a set of the things that they have in their
warehouse. >> Are this available for purchase now?
>> LEGRAND: Think about 11 weeks. Yes, that's what--that's what their target is. It's in
[INDISTINCT] about 20 people have them, and so, you know, it's on the [INDISTINCT] already
>> Then we're going to need--we're going to need processing software that bundled with
it, or is it [INDISTINCT]...? >> LEGRAND: Right, right now, no. And that's,
that's something we're going to add later. Right now we have just the abstraction [INDISTINCT]
that kind of stuff. We're going to add, you know, and we got abstraction library and talk
about it later. Originally we're going to have vision.
>> But we haven't heard [INDISTINCT] the Eclipse? >> LEGRAND: Right.
>> So if I want [INDISTINCT] if you use this as an elective in the school, like a [INDISTINCT],
you know, [INDISTINCT] you have lesson plans and you have [INDISTINCT]?
>> NOURBARKHSH: Yes, you know, go to robot. I think robotdiaries.org or communityrobotics.org,
you should be only get to at least contact information for the people who have a curriculum
[INDISTINCT]. >> Okay.
>> What's' going to be costing [INDISTINCT] for?
>> LEGRAND: Three hundred dollars. We want to bring it down, yes.
>> NOURBARKHSH: I think we're going to do that with volume, that's the thing you catch
on. >> LEGRAND: Yes. But the new one that uses
the Cortex A has six X combination pattern and it may actually have been cheaper.
>> What if I don't want that huge broadside so I'd say I just want to build a little,
you know, tiny robot, you plan to make a version of this but geared towards sockets?
>> LEGRAND: You're killing me. We tried, well, I mean, one thing is I mean, we tried to,
we tried to space optimize it as much as possible without adding much cost and so that's the
product we ended up with. So I think in the future, you know, things will just kind of
naturally tend to shrink, they also... >> [INDISTINCT] to wait.
>> LEGRAND: Yes, I also well, we'll have, you know, we'll have different versions that
maybe don't have as much features and this is kind of a super set of, you know, this'll
satisfy a lot of users we can as offer, you know, different sets of [INDISTINCT.
>> Thank you. >> When will the [INDISTINCT]?
>> LEGRAND: That's like in 2nd quarter of next year. So, it's [INDISTINCT].
>> What kind of meaning [INDISTINCT] you guys building for that stuff but, you know, the
web for PCs doctrine there's kind of work to be done. Will you probably be seeking contributors
to that and are you organizing that effort? >> NOURBAKHSH: That's a good question. We
don't know how organized it gets yet exactly but there's a few things we can do. One is
we can actually use some of the CMU community. We've done this in the past with mini-contest
at CMU where we bring him in for a week or a weekend, give him some challenges, and have
them run, and have gone some excellent example application programs that way. So, probably,
tap our local trade of engineers who don't sleep at night first. But the idea is right.
To create some kind of community that's much bigger than that and that's part of what--I
think we want advice for. >> Is it open source?
>> NOURBAKHSH: Completely. Totally open source. Everything is open. And actually, so let me
ask the question. Raise of hands, how many people would potentially be even interested
in hacking on something on something like this building a robot? We'll that's good.
>> [INDISTINCT]. You have the right audience here?
>> NOURBAKHSH: Yeah. There's some selection bias I guess for the advertising poster.
>> Right. Do you have robotics build in Google? >> NOURBAKHSH: We do.
>> Robots at google.com. So how many members do we have?
>> It's probably a dozen contributors and another dozen workers. We'll just [INDISTINCT].
>> And today, you get together regularly or pack and code. And we get together on Mondays.
And we're trying to do bringing Google apps basically, cloud robotics apps, [INDISTINCT]
occasion on that recognition to the robot. >> Have you scattered us on a platform [INDISTINCT]?
>> No, and we have like half of dozen in play and the best [INDISTINCT]. So we would love
to join like a conversation when you're talking about standardizing your command. We standardized
a few and it's really cool when we could pull out a cell phone and somebody's brand new
robot could fire up, and then they say, "Well, just pair it with them [INDISTINCT] you can
drive it using the code you already have but we only have a little bit standardized but
it goes a lot more. >> NOURBAKHSH: So, what a conversation, that
sounds fantastic. You have a lot of [INDISTINCT] robotics section down the staff of Google
[INDISTINCT]. >> [INDISTINCT].
>> NOURBAKHSH: Plenty of people from [INDISTINCT] too. Well, thanks for the attention.
>> Thank you. >> Thank you.