October 14, 2024 • 46 mins
Tesla showed off the Optimus humanoid robots recently. What's the history behind bipedal, humanoid robots? Why are they so hard to get right? Why would you want to design such a robot in the first place? We explore the history, science and evolution of humanoid robots.
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:04):
Welcome to Tech Stuff, a production from iHeartRadio. Hey therein
Welcome to Tech Stuff, I'm your host, Jonathan Strickland. I'm
an executive producer with iHeart Podcasts and how the tech
are you? So? Last week, Tesla held an event called
(00:24):
we Robot, in which attendees got to see a new
vehicle that was dubbed the robo Van, although Elon Musk
pronounced it as reboven. There was a cyber cab that
must claims is going to cost less than thirty thousand
dollars when it goes into production sometime before twenty twenty seven,
which would potentially allow the average person to run a
(00:47):
small uber or lift business out of their own garage
with the cybercab giving people rides. Though that raises a
lot of questions I have, like liability issues. Let's say
that your cybercab got into an act extent. Are you
held liable for that? Or is Tesla or I don't know.
That's a discussion for another time. But they also gave
(01:07):
a closer look at the humanoid Optimus robots, and they
had robots that were dancing and serving drinks and some
that even held conversations with attendees. Now, those robots had
some help. Tesla also did not hide this fact. This
is not like a gotcha because the company was very
forthright about this. Remote operators were augmenting the robot's abilities
(01:30):
while those robots were on the floor. So, for example,
those conversations were actually human beings who were using the
robots kind of like an advanced bipedal intercom system. But
it made me think about the long history of humans
trying to make humanoid robots. Now, in some ways this
pursuit is a bit strange because a legged, bipedal humanoid
(01:56):
robot brings with it a ton of challenges that you
could sidestep if you just made some compromises to your
design approach, like why does the robot need to be
bipedal and humanoid? If you decided the robot should move
around on wheels or tank treads, or maybe move around
on all fours, or maybe you make like a centaur
(02:19):
like robot where it has like a base with four
legs and then a torso with two arms and it
stands upright. You know, we make the rules like it
doesn't have to take any specific form factor, and you
could do that and get around a lot of the
challenges you would face if you were to instead focus
on this humanoid bipedal approach. Creating a robot that can
(02:42):
move the way humans do is hard. It has taken
decades of research and development to accomplish that to a
reliable degree, and even then it's typically under very controlled circumstances.
When you start getting into stuff like uneven terrain, it
gets a lot trickier. Okay, In other ways, the desire
(03:04):
to build a human like robot is totally understandable. You know,
my first reaction to anyone talking about developing a humanoid
robot is why why are you doing that? What's your reasoning?
Because if you can accomplish the same goal using a
different method of locomotion, that might be the better choice. However,
(03:24):
if you want this robot to be able to do
tasks in our human world, like tasks that human beings
would typically carry out on their own, well, making the
robot human shaped makes more sense. You don't have to
adapt your environment to the abilities of the robot, right
because a set of stairs or a ladder would defeat
(03:47):
most wheeled robots. They would come to it and say, well,
I can't navigate up this obstacle, and if my goal
is on the other end of that obstacle, that's a problem.
So to really maneuver within the human world, it helps
to have your typical human shape and typical human capabilities. Now,
(04:08):
the flip side of this is that we humans, we
have an obligation to make spaces accessible for those who
have an atypical shape or atypical capabilities. We need to
make sure that people who don't have the use of say,
their legs, can still access really important stuff. That's a
responsibility we have, but that's a topic for another episode. Plus,
(04:30):
I think there is something inherent within us, or at
least inherent within some of us, some human beings, and
it drives us to want to create companions that look
and to an extent behave the way we do. So
much of science fiction is based around variations of this idea.
(04:52):
Arguably Mary Shelley's Frankenstein is kind of along this track
of thinking. You know, you have you your crazy scientists
who desires to play god, and then you have General
Capex's influential work, Rossum's Universal Robots. That's the work where
we get the term robot in the first place, or
(05:12):
the replicants in Blade Runner, which actually pretty closely resemble
the robots in capex work, to the works of Isaac
Asimov and beyond. So in many of these pieces there's
a warning that is given that the pursuit to build
and exploit robots often comes tinged with arrogance and hubris,
(05:36):
and it rarely works out well for anybody by the
end of the story. Now, not all of those robots
were mechanical or electro mechanical or digital creatures. In fact,
like Frankenstein's Monster, the robots in Capex's work and the
replicants in Blade Runner are all synthetic life forms. They're
(05:56):
not masses of wires and circuit boards. Asimov's work introduced
more mechanical and electro mechanical creatures with artificial brains. But ultimately,
all the stories involve creatures and creations that gain an
awareness of themselves and their place in the world, and
how they reject the hand that has been dealt to them,
(06:18):
often to dramatic and catastrophic degrees. Now, what the future
holds as far as humanoid robots go has yet to
be written, though we can certainly look back and talk
about some of the history in the field of humanoid
robots that have happened so far. Before robots, we have
examples of automata that mimiced human movement and capabilities. Some
(06:41):
of these were actual clockwork creations, such as the Karakuri.
These are puppets from Japan, dating as far back as
the seventeenth century, so the sixteen hundreds. These used mechanisms
to power certain movements, usually repeatable movements like serving tea
or playing musical instrument. Some of the early automata weren't
(07:04):
automata at all. They were hoaxes. The famed mechanical Turk
creation of Wolfgang von Kemplan is such an example. Kempland
actually stowed a human operator inside a cabinet that was
attached to this supposed automaton that had been designed to
play chess. In reality, it was the human operator hidden
(07:25):
inside the cabinet that was controlling everything. It was really
just a puppet, not an automaton. But these creations, whether
actual automata or not, were limited in function, and typically
they weren't bipedal either, not truly like they weren't moving
around on two legs. They might be stationary in the
case of the mechanical turk, or they might have wheels
(07:47):
and they have like robes that cover up the wheels,
so it looks like they're kind of gliding across, but
they're not actually walking. And I also have to include
one story I came across because it's just too absurd
to leave out. So in eighteen forty eight. In May
of eighteen forty eight, a couple of different journals, one
(08:10):
of them being Scientific American, published an account of a
supposed encounter with a remarkable automaton that was capable of
standing up, sitting down, and even of speaking. The automaton
itself was apparently almost indistinguishable from a human being. The
account was said to have originally published in quote an
(08:32):
Augsburg Gazette end quote. So Augsburg is a city in
Bavaria in Germany, and so the original article was supposedly
written in German and then translated into English to be
published in various periodicals in the United States and elsewhere.
The automaton's name was mister Eisenbrass, which is great, but
(08:54):
the name of the inventor was doctor Lube, which, y'all,
that's like a gift from the gods of comedy right there.
Mister Eisenbrass and Doctor Lube. I maintain that someone should
make a stage production that has the title mister Eisenbrass
and Doctor Lube, and it might well be me unless
someone beats me to it. And someone probably will beat
(09:15):
me to it, because I'm infamous for coming up with
ideas for shows or novels or whatever and then just
sitting on them. But my goodness, mister Eisenbrass and Doctor Lube.
That's that's a title, y'all. Anyway, according to this article,
the author and some other visitors went to the lab
of doctor Lube, which I imagine was down a slippery slope. Anyway,
(09:38):
The doctor was quote seated at a sort of cabinet
having a keyboard somewhat similar to that of a piano
forte arranged on one side of it, and nearly in
the center of a room sat a fashionably dressed gentleman
who rose and bowed as we entered endo quote. And then,
according to the article, the visitors engaged in some small
(10:00):
talk with this fashionably dressed gentleman, and the gentleman actually
took a seat after the visitors had sat down, and
eventually the doctor stops playing at his keyboard, and mister
Eisenbrass goes quiet, and Lube explains that the whole thing
is a mechanical contraption. Only then do the visitors notice
the cables going from the keyboard console to the chair
(10:23):
of mister Eisenbrass. Now, according to the piece, Lube procured
bones from a human being, presumably a dead one, which
is a big ol' ick already, but then use rubber
tubes to kind of serve as musculature, and that he
also created quote a perfect system of nerves made of
(10:43):
fine platinum wire and covered with silk end Quote to
what end, you might say, what what are the nerves for?
I guess the idea is that electric motors would pull
upon the rubber tubes. It does explain that there were
electro magnets that were in use in this system, and
that the tubes just served as muscles that when you
(11:05):
pulled on them, would cause the contraction you would associate
with a human body. I actually at first assumed, since
they were talking about rubber tubes, that this was going
to be a pneumatic system where you would use air
to achieve similar results. Right, you pump air into something
to extend a limb, and you would allow air to
escape to contract the limb. But that apparently is not
(11:27):
how this was supposed to work. Supposedly, the keyboard allowed
the doctor to produce incredible results just by pressing a
few keys, like I guess there was a key that
was just labeled small talk or something, and that the
figure was apparently capable of quote walking, talking, singing, playing
the piano, and doing many other things with as much
(11:49):
ease and precision as an accomplished man. End quote. The
author then proactively chides the reader for undoubtedly asking so
what good is on this? And then the author goes
on to talk about how mechanical servants will replace all
those undependable lauts and scally wags who currently act as servants,
and thus give the women of the household the freedom
(12:12):
to carry out their feminine duties as caretaker of the
home just by tickling some ivories. Yeah, this article is
well and truly both sexist and classist. Anyway, to say
that I am skeptical about this account is putting it lightly.
I feel fairly certain that no such demonstration ever actually
(12:33):
took place, or if there were some kind of demonstration,
it did not unfold as described in this article. I
did try to find the original article written in German.
I assumed that the German gazette that was referenced in
the English piece must have been the Algemina Zeitung that
was published in Augsburg, Germany for most of the nineteenth century,
(12:56):
and was like the main paper not just of Augsburg,
but like of that region of Germany. However, I found
no record of Eisenbrass or doctor Lube anyway. As diverting
as mister Eisenbrass and doctor Lube are, I feel confident
in saying that the capability of building a robot that
could maintain its balance standing still, let alone walking around
(13:21):
all without other means of support, was likely well outside
the reach of even the most clever of innovators in
the mid eighteen hundreds. That seems like that's just a
no brainer. When it comes to creating a two legged robot,
one where most of the mass is actually above the
legs of the robot, not contained within the legs, things
(13:42):
get really tricky because a lot of physics have to
be considered before engineers could get serious about bipedal robots.
If we relied solely on trial and error and just
figured we're going to get this right, we'd likely not
be anywhere as far along as we are right now.
So when we come back, we're going to consider how
(14:03):
challenging it is to create something that walks around on
two legs. But before we get to that, and I
get a little unbalanced, let's take a quick break to
thank our sponsors. All right, So what's the big deal
(14:27):
with walking around on two legs? Lots of people do
it all the time. You know, toddlers can get the
hang of it without too much trouble, and we celebrate
it when it happens like that's a big deal, But
then shortly after that it's It really becomes just a
source of stress as the toddlers toddle along toward one
danger or another. But you're entirely dependent upon just two
(14:51):
points of contact with your surrounding environment. If you're talking
about a true bipedal form that is capable of moving
around round the area, and those two points of contact
with the environment are essentially the bottoms of the foot seats. That's,
of course, if everything is actually going as you wanted
it to. If things have gone poorly, you might actually
(15:12):
have lots of different points of contact with the ground.
But that's because you went horizontal after something went wrong.
So you've got your robot. It's got two legs, the
two points of contact with the ground or the bottom
of the feet likely the legs and the feet, and
your robot in general has a number of degrees of freedom.
So degrees of freedom are joints that allow movement along
(15:34):
at least one axis. The point of contact with the
ground could actually be considered a passive degree of freedom
in itself. And you're also relying on friction to allow
your robot to stand up, to maintain balance, and to
get anywhere. If the robot's feet were frictionless, then it
wouldn't be able to stay up right at all, let
(15:54):
alone walk. And you've got all this weight above the
legs that you have to worry about. So have you
ever balanced something like, say, a baseball bat, on the
palm of your hand. If you do that with a
baseball bat and you're using the narrow part the handle
in of the words of the bat and you're balancing
that on your palm and the thick part of the
(16:15):
bat is up in the air, you know that little
motions can create big results. Right A small movement at
the base can cause the top to really sway, and
that requires you to make larger corrections down at the
base in order to keep everything in balance. Well, that's
kind of what it's like to try and figure out
how to make a bipedal robot walk while maintaining its balance.
(16:37):
You've got a inertia to deal with, and that really
affects balance. How do you make sure the mass above
the legs doesn't throw everything off kilter whenever it starts
moving or when it stops moving. How do you correct
for that so that your robot doesn't just tumble over?
How do you keep your bot upright? One early discussion
(16:57):
that became a core component in the pursuit of bipedal
robots revolved around a concept that came to be known
as the zero moment point or ZMP. So a pair
of smarty pants is from Russia named Beyomir Vukabratovich and
Davor Jurichic first described this back in the nineteen sixties. Now,
(17:21):
the actual term zero moment point would be coined a
little bit later, but it was used to describe what
they were talking about, and the whole concept revolves around
a moment in which the net reaction forces between a
bipedal mechanism's feet and the ground are essentially zero and
there's no movement along the horizontal plane. So that means,
(17:45):
like forward momentum and gravity they've kind of canceled each
other out. You've hit this zero moment where you don't
have to worry about the robot tipping forward and falling over.
It is you could think of as a moment of stability,
and the robot, assuming no external forces are acting upon it,
will remain upright, assuming that ZMP is maintained. Now, this
(18:09):
discussion really drives home how stability can be a huge challenge.
As robots move, they must deal with inertia and you
have to know the math to achieve dynamic stability if
your robot is to remain upright, whether it's walking, running, jumping,
or whatever. And in fact, it gets increasingly more difficult
as you go down those tasks. Walking is hard, running
(18:33):
is really hard, and jumping is really really hard. Like
the jump part might be easy, the landing and staying
upright that's the hard part. Otherwise your robot's gonna topple over.
And while watching a robot take a tumble might be
a great YouTube video. In practice, in the real world,
you obviously don't want your robots to be falling over.
You want your robots to be stable and capable of
(18:55):
moving around environments without causing damage or potentially injury to people,
to be able to maneuver. It's tough and robots are expensive.
You don't want them falling over. Then you're thinking, well,
that's twenty thousand dollars to get this thing back on
its feet. Again, that's not a great way to reach
progress either. So to plan the motion for a robot,
(19:18):
you need to be able to calculate the zero moment
point or ZMP. You need to figure out which joints
the robot is going to have to engage in order
to achieve stability under its various operating conditions, and those
conditions could include things outside of your strict control. It's
one thing to calculate how the robot can achieve stability
(19:38):
when it's walking across a level floor that has a
firm surface, But what about a floor that surface isn't
as firm. Maybe it's a little squishy, you know, maybe
it's a nineteen seventies shag carpet or something, or what
if the floor isn't level, what if the terrain is
actually uneven, you know, kind of like a typical sidewalk
in the city of Atlanta. How does a robot compensate
(20:01):
for all this, remain stable and keep itself from pitching over?
This is a non trivial challenge, and it takes a
lot of work to get to a point where robots
are sophisticated enough to achieve stability. Engineers have had to
take a lot into consideration. Would more degrees of freedom
help or does that actually overcomplicate matters? I mean, there's
(20:23):
no reason why we should be constrained to the same
degrees of freedom that a person has. Right Like, we
could think, oh, let's mimic the way humans work and
make sure that the ankles and the knees and the
hips all have the same points of articulation. Or we
could say, well, there's no reason why we couldn't have
more or fewer joints if it makes the operations work better.
(20:47):
So that's something to think about, you know. And then
walking would require shifting stability so that the robots can
maintain itself with just one foot in contact with the ground,
right like, suddenly points of contact have halved. If you're running,
it's even harder because with running, at least the definition
of running that roboticists use, there's a moment that may
(21:10):
only last just for a split second, but there's a
moment in which both of the robots feet are not
touching the ground. So how do you achieve stability when
your point of contact is continually interrupted? Jumping is even harder, right,
because you're really leaving the ground then, and how do
you ensure that landing you do so in a way
(21:32):
that you maintain stability. So that ZMP was a huge deal.
It still is a huge deal. Not all robot locomotion
is centered around ZMP calculations, by the way, but a
lot of it still is. So a lot of the
work in bipedal robots, particularly in the seventies and eighties,
involved calculating ZMP and keeping the robot within that zone
(21:52):
while engineering movements. So calculating ZMP is one thing. Building
a robot that can balance is another. Even building robot
that's capable of static stability is no simple task. Static
stability just means the robot is able to stand still
and not fall over, and believe it or not, that's
easier said than done. In nineteen seventy four, labs at
(22:14):
the School of Science and Engineering at Waseda University in
Japan started a project with the goal of building a
stable bipedal robot. This project became known as Weibot wabot
that stands for Waseda Robot. The labs created a new
focus group within the laboratory system at the university. It's
(22:36):
called the Bioengineering Group, and their first effort, the Waybot one,
wasn't exactly something that you would mistake as a human being.
It was not like a smooth humanoid robot. You would
never look at it and think it was anything other
than a robot. In fact, it looked kind of like
a giant erector set. It was very much a big, blocky,
(22:57):
metallic humanoid robot. So it had a limb control system,
obviously very important if you're going to have a walking robot.
It also had a primitive vision system. It even was
able to converse in Japanese to a certain degree. They
said that it had the in electrical capacity of like
a one and a half year old Keep in mind
this is the early nineteen seventies. It was tethered to
(23:20):
computer systems and power systems. There was not yet a
point where we had reached a miniaturization where you could
have all that computing power, not to mention electrical power
on board the robot itself. If you had done that,
the robot would either need to be huge or would
be carrying the biggest backpack you've ever seen in order
to have all the computational and electrical power necessary to
(23:41):
operate this thing. So, yeah, it was tethered. The Waybot
was the first humanoid robot to achieve static stability, and
later it would be the first digitally controlled anthropomorphic robot
at all that was able to achieve dynamic stability, and
statically stable means that the robot can remain balanced while
(24:03):
standing still. Dynamically balanced refers to robots that are maintaining
that stability while they are still in motion. Going from
one to the other and back again is actually really hard.
Like achieving one is hard, achieving the other one is
also hard. Going back and forth between still and moving
and maintaining balance is even harder. You know, remaining dynamically
(24:25):
stable is often easier than going from dynamic to static,
assuming that you know motions are smooth, and fast enough
to counter the forces acting on the robot, because when
you think about it, walking is really a series of falls.
It's like you're falling and you're catching yourself over and
(24:45):
over again. You move forward when you walk, not just
by moving your legs. You know, you kind of lean
forward into the walk as well, and for a moment
it's as if you're falling forward and that you would
face plant into the ground ahead of you. But of
course you've moved your leg to catch yourself, and this
process repeats, so you're propelling yourself forward by constantly almost
(25:08):
falling but catching yourself over and over and over again.
Now we do this without really thinking about it, like
once we learn how to walk, we don't have to
focus on this. This is this is just how we
do it. Robots, however, have to calculate this stuff in
order to do it properly, in order to catch themselves
with just the right amount of force to keep things
moving and not to use too much or too little
(25:30):
force and then risk taking a fall. Tons of work
in robotics continued around the world towards this goal of
creating bipedal humanoid robots, but nearly all the articles I
read for this episode. Make a big jump from the
nineteen seventies to the mid nineteen nineties. That's when Honda
unveiled a robot that they called the humanoid P two.
(25:53):
So Honda had actually been developing humanoid robots for about
a decade before showing off the P two. A series
that was kind of the in secret R and D
works that was called the E series of robots, and
this was in the nineteen eighties. These E robots were primitive,
but they demonstrate the capability to walk on level ground
(26:15):
under very controlled circumstances. The first E robot, for example,
was able to walk at an extremely deliberate pace because
it was said that each step took about five seconds
to complete. I challenge you to try walking that way,
even if it's just a couple of feet, count to
five slowly per step. That is slow, but it illustrates
(26:39):
how challenging it was to design a robot capable of
walking in a biped away. Now, the first E series
robot was built in nineteen eighty six. It essentially looked
like a pair of legs attached to robotic hips and
that's it like. It didn't have any top half. There
was no torso or arms or anything like that. The
B two wouldn't debut until nineteen ninety six. Honda had
(27:00):
a decade of work developing humanoid robotics before they revealed
to the public what they had been up to. And
like I said, E zero, not very humanoid. It is
like a free, free walking pelvis, robotic pelvis. The later
robots in the E series started to look I don't know,
in my opinion, even more strange, Like one of them
(27:22):
looks kind of like a microwave oven that has legs.
One of them looks sort of like the front end
of a fancy car, like a Rolls Royce, but with legs.
You would never call any of them human in shape,
but they were gradually evolving toward that. The P two
in nineteen ninety six looked much more human in that
(27:44):
it had legs, and it had arms and a torso
and a head. Now this head was rectangular, and it
was it was wider than it was tall. It was
not human looking at all, but it demonstrated that Honda
had been hard at work tackling this BikeE Hetle challenge,
and it would serve as a foundation for a much
more famous Honda robot just a few years later, a
(28:07):
robot that would debut in two thousand. It was called Asmo.
Now I have a fun connection with Asimo, and I
will talk about that, but first let's take another quick break.
(28:29):
All right, So the year was two thousand and seven.
Asimo had been a thing for the better part of
the two thousands, like it debuted in early two thousand
and now it's two thousand and seven. I had just
been hired by a company called HowStuffWorks dot com, and
one of my first assignments was to rewrite and to
(28:49):
update the article on how Asimo works. So I researched
the project, including the lofty goal that the engineers had
set to have a robot that couldn't just walk but
could also run. So it would be a robot that,
at little points would have both feet leave the ground
just for a moment, but still be able to maintain
(29:10):
its balance. Now that was a huge accomplishment. Even if
you were to watch videos and it kind of looks
like Asmo's doing a little hopping dance you might do
if you were in need of getting to a restroom,
Like I think of it as oh, it's doing the
peepee dance. But you know, watch Asmo running. It's cute,
it's weird, it has this odd sort of tone to it,
(29:30):
I would say, But it was phenomenal because yes, for
a split second, both feet are off the ground, and
yet when the opposite foot makes contact with the floor,
the robot would maintain its balance and be able to
continue running. Asimo looks a lot more human than P
two did. In fact, it looks like sort of a
(29:51):
diminutive astronaut in a spacesuit. I actually got to meet
Asimo once when I was at Disneyland in California, because
there was demonstration of Asimo that was a real blast.
I watched this presentation that they gave, and then I
mentioned to a cast member that I had written an
article about how Asima worked, and they brought me aside
and I got to meet the robot, which was a
really fun moment for me. That was kind of cool. Anyway,
(30:15):
Asimo would go on to establish a lot of firsts
in the bipedal humanoid robot space. Not only was it
the first one to run, it could also climb and
descend stairs, at least eventually it could. One early demonstration
didn't go so well, and Asimo tripped, but it later
demonstrated those capabilities, and these were things that would be
(30:38):
built upon in future robotic projects, both at Honda and elsewhere.
I should also mention that Asimo was largely a programmed robot,
in that it would maneuver around and interact with an
environment that had been carefully mapped out for the robot.
So it's not like it was spontaneous. It wasn't coming
into a brand new environment finding its way or around
(31:00):
picking things up that kind of stuff. It was following
a very specific set of instructions and it knew where
everything was supposed to be and where it was supposed
to go. So it was not like an autonomous robot.
But that really wasn't what the project was about. It
wasn't about autonomy. It was about robotic locomotion and interacting
(31:21):
in human spaces. So you have to keep in mind
that this whole approach to robotics is multidisciplinary in nature.
It requires lots of different work in varying fields, some
of which aren't even in technology. I'll talk more about
that later. In twenty fifteen, the Defense Advanced Research Projects
(31:44):
Agency here in the United States aka DARPA held a
robotics challenge. So DARPA is known for contracting with various
companies and research facilities to develop bleeding edge technologies potentially
useful for the purposes of national defense. They're not always
couched in those specific terms, but that's the directive of DARPA.
(32:05):
DARPA has played a part in everything from the development
of the Internet to the early days of driverlest car technology.
Well in twenty fifteen, they had a lofty goal set
for robotics teams and it was all inspired by a
terrible disaster. So in twenty eleven, an earthquake and tsunami
damaged the Fukushima Daichi Nuclear Power Plant in Japan. The
(32:28):
power plants backup systems were damaged, and this led to
a situation in which the plant's reactors began to overheat
because there was no power that could be used to
operate the cooling system in order to keep everything under control,
and ultimately this overheating led to a containment failure and
radioactive material was released into the environment. It was the
(32:49):
worst nuclear disaster since Chernobyl. Cleaning up after the disaster
was a really dangerous job. Response workers would be subjected
to potentially dangerous levels of radiation or extended periods of time.
DARPA's challenge was to give robotics teams a set of
tasks that a robot would have to complete with minimal
direction and intervention from the teams. The idea being let's
(33:12):
work toward a technological solution where we could develop robots
that could step in into situations that were like Fukushima
and take the place of humans so that human beings
don't put their lives at risk to do this kind
of stuff. The robots, since they have no lives, they
could go and do it and we wouldn't be putting
any human life in jeopardy. That was the concept. But
(33:34):
to do this, these robots would have to do very
human like things, and they have to maneuver in a
very human like world because it was designed by humans, right,
so no big surprise there. So the robots would have
to do stuff like get in, to get out of
and operate a vehicle, to be able to open doors
and step through doorways, to pick up a tool and
(33:56):
to use it properly. And teams were given some but
not all, the information that they would actually need in
order to complete the various tasks that DARPA had laid out,
and the reason why they weren't given everything is because
the whole concept requires teams to build a robot that
could accomplish goals in a world that is unpredictable and
uncontrolled for the most part. In a real emergency, there
(34:19):
may be no way to account for all the variables,
and depending on the nature of the emergency, a team
might not be able to maintain a direct connection with
their robot, so the robot would need to be able
to handle some of this autonomously. Now, most of the
challenges were pretty darn straightforward, and they would have been
trivial for most people to be able to complete if
(34:41):
they were given the assignment. You know, if you ask
your typical human to get into a vehicle kind of
like a golf cart and to drive to a specific location,
to then get out of that vehicle, to open a door,
go through the door, pick up a power drill, drill
a hole in a wall, climb some stairs, and navigate
some rubble, and you'd likely see a lot of people
(35:01):
succeed at this. It's again a pretty simple set of
tasks for most people, that's not a tough gig, but
for robots, it's a totally different story. Now. There are
compilations of videos of the various teams participating in this
competition that show just how challenging it really was. There
are videos of robots that, upon trying to just walk
through a doorway, fall over. One robot completed the list
(35:27):
of tasks, turned wave to the crowd, then fell over
because that balance thing is really hard, y'all. Creating humanoid
robots that can interact with the technology that was made
for human beings requires a ton of consideration and cross
disciplinary work. For one thing, sometimes it requires roboticists to ask, hey,
(35:48):
why did we make this thing work in this specific way.
It's almost like you have to go through reverse engineering
the world around you in order to understand why things
are the way they are. That doesn't always end with
an answer that makes much sense. By the way, sometimes
we're like, wow, we should really change this because this
is not the best way to do it, but at
that point it's the established way to do it. Ultimately,
(36:10):
a team from the Republic of Korea won this competition.
The winning robot was named DRC dash Qubo Hubo. It
completed the series of tasks in just under forty five minutes,
and the team took home a two million dollar cash prize. Now, no, lie,
two million dollars, that's a lot of money, But I
(36:31):
would be willing to bet that the two million dollars
doesn't remotely cover the costs of all the research, development,
and production of the robot itself. I bet if you
were to add up all the expenses of making this robot,
it would be more than two million dollars. But the
money wasn't really the full goal of this thing. Like
that was an award, but you weren't doing it to
(36:53):
win the money. It's this challenge trying to figure out
a way to achieve this really tough goal set out
by the nature of the challenge itself. That's the real
call the engineers out there, know what I'm talking about.
Like that thrill of tackling a problem and figuring out
a solution. That's really what drives a lot of engineers. Now,
(37:13):
if anything, the challenge illustrated just how hard it is
to build a humanoid robot that can function properly. Now,
the benefits are pretty clear. You know, this kind of
robot could potentially step in during situations like the Fukushima
disaster scenarios in which a human would be put into
danger and the robot, due to its design, could interface
with systems that had been built for humans. That's understandably
(37:35):
a worthy goal. It's just a very challenging one. And
it gets harder when we start to bring artificial intelligence
into this because we've been mostly focused on things like locomotion.
But let's talk about AI. Now. I've told this story before,
but I went to a panel and about robotics. It
was at south By Southwest. This was several years ago,
and at that panel, the presenters were talking about how
(37:57):
challenging it is to teach robots how to do things,
like not program the robots to do it, but to
teach them how to learn in an environment and then
replicate things that they have learned. Like even when you
build models in which the robots are able to observe
and then attempt to replicate actions, stuff can go wrong.
Robots have the same limitations as other examples of machine learning. So,
(38:22):
for example, I have often used an example saying, like
teaching a computer to do something like recognize that an
image represents a specific object like a coffee mug is hard.
Not all coffee mugs are alike, right, Some come in
different sizes or shapes or colors. Some might have handles.
Those handles could be shaped one way versus another, some
(38:42):
might be you know, the pictures of them might be
in under different lighting conditions, or they might be paired
with other stuff that's of a similar size or shape
to the coffee mug. All of these elements represent challenging
variables to machines that are being trained to recognize images.
You know, the machines do not inherently understand what makes
(39:03):
a coffee mug a coffee mug. That's what you are
teaching them. And you know, you can teach a human
being what a coffee mug is and they pretty much
get it pretty darn quickly, even to the point where
they can recognize other coffee mugs that don't look exactly
like the initial one. But for computers it requires a
lot more training. You train your model, then you tweak
all the settings so that you can improve your results,
(39:25):
you know, cut down on the false positives and fix
all the mistakes, and you train it again, and you're
potentially using millions of images in order to do this. Now,
consider the humble door, now a door is a pretty
darn simple thing to operate for most people, but there
are lots of different ways that a door could potentially
operate right Like, the door might have a knob that
(39:47):
you have to twist. It might have a bar that
you have to press, or a handle that you have
to pull. So when you encounter a door, chances are
you have a decent idea of how it works, but
you might not know which way it opens it for
that can give you a little bit of a pause.
One of my favorite of Gary Larson's Far Side cartoons
a classic cartoon strip from the eighties. It shows a
(40:09):
young boy pushing as hard as he possibly can on
a door, and just above his hand on the door
is a label that reads pull, and next to the
doorway is a sign that reads Midvale School for the Gifted.
I feel this cartoon in my soul some days. Well.
Robots are kind of like that student. Even a robot
that's been trained to open doors might need to pause
(40:30):
and have a digital think about things before giving it
an attempt. So the south By Southwest panelist was telling
the story of such a robot, and this robot sat
outside a door in a hallway. I think It was
an electrical engineering department at the university that this panelist
worked at, and it was sat there for several days
just contemplating the door, and it was really irritating folks
(40:53):
who worked there because they had to walk around this
thing in order to get through the hallway, and if
they passed in front of it, it irritated the row
bodicists because it could actually disrupt the process and set
things back even further. But the robot was just trying
to figure out how it should proceed in order to
try and open the door. And when you think about
how a robot could potentially be powerful enough to cause
(41:14):
damage to the environment it's in if it attempts to
do something incorrectly, then you start to understand why taking
time might actually be a necessity. It might be an
important thing to build into robots. It seems ridiculous to
just stare at a door for days on end before
even attempting to open it, but if you could potentially
(41:34):
rip the handle off the door or damage the door
in some way, well, taking time might be a needed precaution.
And that's not even getting into the challenge of having
robots that operate within an environment in which they're also
human beings. Obviously, you have to take a lot into
consideration in those kinds of situations where robots and human
(41:55):
beings are going to be working within the same environment.
Like in most industrial uses for robots, the robots are
very much separated from all the people, like there are
multiple safety considerations put in place to keep the robots
and people away from each other because the potential for
catastrophe is way too high. If you're working too close
(42:16):
to a robot, like it's a robot that welds stuff
or whatever. Well, you know you don't want to get
in the way of a welding, right, That would be awful,
It would potentially be deadly. So creating robots that are
capable of interacting among human beings it comes with its
own series of challenges you have to overcome. So not
only must the robot be able to move around without
(42:37):
falling over onto somebody, it needs to be able to
do this in a way that doesn't cause anxiety or
fear or other negative reactions among the human beings. Really
weird thing is that sometimes a robot can behave a
little too human and that can end up being almost
as bad as if it's not acting human enough. You've
got to find a balance, Like there are expectations that
humans have when it comes to interacting with the robots.
(43:00):
Of the robots behave too far outside that set of expectations,
it can cause issues. So robotics has become a truly
multidisciplinary endeavor. Making a bipedal humanoid robot capable of integrating
with humans the way the Tesla Optimist robot is supposed
to do that requires lots of work in disciplines that
(43:21):
go well outside of technology. We're talking about stuff like psychology,
and I think every time we see a remarkable achievement
in the robotics space, we're also reminded how far we
still have to go and how hard this really is.
So will Tesla's Optimist robots deliver upon all the promises
that Elon Musk often quotes at these events. Maybe I'm skeptical,
(43:46):
largely because Elon Musk has proven to make some rather
ambitious claims the past that have failed to manifest as described.
He's kind of the boy who called fully autonomous driving.
And largely I also have doubts because I have an
inkling as to how hard it's going to be to
make a bipedal general purpose robot that's at least as
(44:06):
good as, and hopefully better than a human being at
doing your typical tasks. If the robot is worse at
doing those tasks, then it's a waste of time and
money to use the robot. Just hire somebody else to
do it, it makes more sense. So like these are
these are really high barriers that you have to you
have to get over, and I don't think we're going
(44:28):
to get over them very quickly. I think it's going
to take years and years more work. But it is
a heck of a goal to aim for. I don't
want to shame anyone for taking aim at achieving this
really difficult task because it drives innovation. I think that's important.
So I don't want to dismiss anyone who's working toward
(44:49):
building bipedal humanoid general purpose robots that have a level
of AI that allow them to operate autonomously within a
human environment. I think that that is a phenomen I
just think it's also one that's going to require many
more years of work for it to be a viable project. Right, Like,
I guess I could see a disappointing version becoming a
(45:13):
reality within a couple of years. But that's really falling
fall short of the promise, and I would much rather
see more work being done to improve the technology than
for a premature release of some humanoid robot that just
doesn't do anything well enough to justify its existence. That
would really take a lot of wind out of the sales.
(45:36):
I think, all right, that's it for this episode of
tech Stuff. I hope you're all well, and I'll talk
to you again really soon. Tech Stuff is an iHeartRadio production.
For more podcasts from iHeartRadio, visit the iHeartRadio app, Apple Podcasts,
(45:58):
or wherever you listen to your favorite shows.
Welcome to Tech Stuff, a production from iHeartRadio. Hey therein
Welcome to Tech Stuff, I'm your host, Jonathan Strickland. I'm
an executive producer with iHeart Podcasts and how the tech
are you? So? Last week, Tesla held an event called
(00:24):
we Robot, in which attendees got to see a new
vehicle that was dubbed the robo Van, although Elon Musk
pronounced it as reboven. There was a cyber cab that
must claims is going to cost less than thirty thousand
dollars when it goes into production sometime before twenty twenty seven,
which would potentially allow the average person to run a
(00:47):
small uber or lift business out of their own garage
with the cybercab giving people rides. Though that raises a
lot of questions I have, like liability issues. Let's say
that your cybercab got into an act extent. Are you
held liable for that? Or is Tesla or I don't know.
That's a discussion for another time. But they also gave
(01:07):
a closer look at the humanoid Optimus robots, and they
had robots that were dancing and serving drinks and some
that even held conversations with attendees. Now, those robots had
some help. Tesla also did not hide this fact. This
is not like a gotcha because the company was very
forthright about this. Remote operators were augmenting the robot's abilities
(01:30):
while those robots were on the floor. So, for example,
those conversations were actually human beings who were using the
robots kind of like an advanced bipedal intercom system. But
it made me think about the long history of humans
trying to make humanoid robots. Now, in some ways this
pursuit is a bit strange because a legged, bipedal humanoid
(01:56):
robot brings with it a ton of challenges that you
could sidestep if you just made some compromises to your
design approach, like why does the robot need to be
bipedal and humanoid? If you decided the robot should move
around on wheels or tank treads, or maybe move around
on all fours, or maybe you make like a centaur
(02:19):
like robot where it has like a base with four
legs and then a torso with two arms and it
stands upright. You know, we make the rules like it
doesn't have to take any specific form factor, and you
could do that and get around a lot of the
challenges you would face if you were to instead focus
on this humanoid bipedal approach. Creating a robot that can
(02:42):
move the way humans do is hard. It has taken
decades of research and development to accomplish that to a
reliable degree, and even then it's typically under very controlled circumstances.
When you start getting into stuff like uneven terrain, it
gets a lot trickier. Okay, In other ways, the desire
(03:04):
to build a human like robot is totally understandable. You know,
my first reaction to anyone talking about developing a humanoid
robot is why why are you doing that? What's your reasoning?
Because if you can accomplish the same goal using a
different method of locomotion, that might be the better choice. However,
(03:24):
if you want this robot to be able to do
tasks in our human world, like tasks that human beings
would typically carry out on their own, well, making the
robot human shaped makes more sense. You don't have to
adapt your environment to the abilities of the robot, right
because a set of stairs or a ladder would defeat
(03:47):
most wheeled robots. They would come to it and say, well,
I can't navigate up this obstacle, and if my goal
is on the other end of that obstacle, that's a problem.
So to really maneuver within the human world, it helps
to have your typical human shape and typical human capabilities. Now,
(04:08):
the flip side of this is that we humans, we
have an obligation to make spaces accessible for those who
have an atypical shape or atypical capabilities. We need to
make sure that people who don't have the use of say,
their legs, can still access really important stuff. That's a
responsibility we have, but that's a topic for another episode. Plus,
(04:30):
I think there is something inherent within us, or at
least inherent within some of us, some human beings, and
it drives us to want to create companions that look
and to an extent behave the way we do. So
much of science fiction is based around variations of this idea.
(04:52):
Arguably Mary Shelley's Frankenstein is kind of along this track
of thinking. You know, you have you your crazy scientists
who desires to play god, and then you have General
Capex's influential work, Rossum's Universal Robots. That's the work where
we get the term robot in the first place, or
(05:12):
the replicants in Blade Runner, which actually pretty closely resemble
the robots in capex work, to the works of Isaac
Asimov and beyond. So in many of these pieces there's
a warning that is given that the pursuit to build
and exploit robots often comes tinged with arrogance and hubris,
(05:36):
and it rarely works out well for anybody by the
end of the story. Now, not all of those robots
were mechanical or electro mechanical or digital creatures. In fact,
like Frankenstein's Monster, the robots in Capex's work and the
replicants in Blade Runner are all synthetic life forms. They're
(05:56):
not masses of wires and circuit boards. Asimov's work introduced
more mechanical and electro mechanical creatures with artificial brains. But ultimately,
all the stories involve creatures and creations that gain an
awareness of themselves and their place in the world, and
how they reject the hand that has been dealt to them,
(06:18):
often to dramatic and catastrophic degrees. Now, what the future
holds as far as humanoid robots go has yet to
be written, though we can certainly look back and talk
about some of the history in the field of humanoid
robots that have happened so far. Before robots, we have
examples of automata that mimiced human movement and capabilities. Some
(06:41):
of these were actual clockwork creations, such as the Karakuri.
These are puppets from Japan, dating as far back as
the seventeenth century, so the sixteen hundreds. These used mechanisms
to power certain movements, usually repeatable movements like serving tea
or playing musical instrument. Some of the early automata weren't
(07:04):
automata at all. They were hoaxes. The famed mechanical Turk
creation of Wolfgang von Kemplan is such an example. Kempland
actually stowed a human operator inside a cabinet that was
attached to this supposed automaton that had been designed to
play chess. In reality, it was the human operator hidden
(07:25):
inside the cabinet that was controlling everything. It was really
just a puppet, not an automaton. But these creations, whether
actual automata or not, were limited in function, and typically
they weren't bipedal either, not truly like they weren't moving
around on two legs. They might be stationary in the
case of the mechanical turk, or they might have wheels
(07:47):
and they have like robes that cover up the wheels,
so it looks like they're kind of gliding across, but
they're not actually walking. And I also have to include
one story I came across because it's just too absurd
to leave out. So in eighteen forty eight. In May
of eighteen forty eight, a couple of different journals, one
(08:10):
of them being Scientific American, published an account of a
supposed encounter with a remarkable automaton that was capable of
standing up, sitting down, and even of speaking. The automaton
itself was apparently almost indistinguishable from a human being. The
account was said to have originally published in quote an
(08:32):
Augsburg Gazette end quote. So Augsburg is a city in
Bavaria in Germany, and so the original article was supposedly
written in German and then translated into English to be
published in various periodicals in the United States and elsewhere.
The automaton's name was mister Eisenbrass, which is great, but
(08:54):
the name of the inventor was doctor Lube, which, y'all,
that's like a gift from the gods of comedy right there.
Mister Eisenbrass and Doctor Lube. I maintain that someone should
make a stage production that has the title mister Eisenbrass
and Doctor Lube, and it might well be me unless
someone beats me to it. And someone probably will beat
(09:15):
me to it, because I'm infamous for coming up with
ideas for shows or novels or whatever and then just
sitting on them. But my goodness, mister Eisenbrass and Doctor Lube.
That's that's a title, y'all. Anyway, according to this article,
the author and some other visitors went to the lab
of doctor Lube, which I imagine was down a slippery slope. Anyway,
(09:38):
The doctor was quote seated at a sort of cabinet
having a keyboard somewhat similar to that of a piano
forte arranged on one side of it, and nearly in
the center of a room sat a fashionably dressed gentleman
who rose and bowed as we entered endo quote. And then,
according to the article, the visitors engaged in some small
(10:00):
talk with this fashionably dressed gentleman, and the gentleman actually
took a seat after the visitors had sat down, and
eventually the doctor stops playing at his keyboard, and mister
Eisenbrass goes quiet, and Lube explains that the whole thing
is a mechanical contraption. Only then do the visitors notice
the cables going from the keyboard console to the chair
(10:23):
of mister Eisenbrass. Now, according to the piece, Lube procured
bones from a human being, presumably a dead one, which
is a big ol' ick already, but then use rubber
tubes to kind of serve as musculature, and that he
also created quote a perfect system of nerves made of
(10:43):
fine platinum wire and covered with silk end Quote to
what end, you might say, what what are the nerves for?
I guess the idea is that electric motors would pull
upon the rubber tubes. It does explain that there were
electro magnets that were in use in this system, and
that the tubes just served as muscles that when you
(11:05):
pulled on them, would cause the contraction you would associate
with a human body. I actually at first assumed, since
they were talking about rubber tubes, that this was going
to be a pneumatic system where you would use air
to achieve similar results. Right, you pump air into something
to extend a limb, and you would allow air to
escape to contract the limb. But that apparently is not
(11:27):
how this was supposed to work. Supposedly, the keyboard allowed
the doctor to produce incredible results just by pressing a
few keys, like I guess there was a key that
was just labeled small talk or something, and that the
figure was apparently capable of quote walking, talking, singing, playing
the piano, and doing many other things with as much
(11:49):
ease and precision as an accomplished man. End quote. The
author then proactively chides the reader for undoubtedly asking so
what good is on this? And then the author goes
on to talk about how mechanical servants will replace all
those undependable lauts and scally wags who currently act as servants,
and thus give the women of the household the freedom
(12:12):
to carry out their feminine duties as caretaker of the
home just by tickling some ivories. Yeah, this article is
well and truly both sexist and classist. Anyway, to say
that I am skeptical about this account is putting it lightly.
I feel fairly certain that no such demonstration ever actually
(12:33):
took place, or if there were some kind of demonstration,
it did not unfold as described in this article. I
did try to find the original article written in German.
I assumed that the German gazette that was referenced in
the English piece must have been the Algemina Zeitung that
was published in Augsburg, Germany for most of the nineteenth century,
(12:56):
and was like the main paper not just of Augsburg,
but like of that region of Germany. However, I found
no record of Eisenbrass or doctor Lube anyway. As diverting
as mister Eisenbrass and doctor Lube are, I feel confident
in saying that the capability of building a robot that
could maintain its balance standing still, let alone walking around
(13:21):
all without other means of support, was likely well outside
the reach of even the most clever of innovators in
the mid eighteen hundreds. That seems like that's just a
no brainer. When it comes to creating a two legged robot,
one where most of the mass is actually above the
legs of the robot, not contained within the legs, things
(13:42):
get really tricky because a lot of physics have to
be considered before engineers could get serious about bipedal robots.
If we relied solely on trial and error and just
figured we're going to get this right, we'd likely not
be anywhere as far along as we are right now.
So when we come back, we're going to consider how
(14:03):
challenging it is to create something that walks around on
two legs. But before we get to that, and I
get a little unbalanced, let's take a quick break to
thank our sponsors. All right, So what's the big deal
(14:27):
with walking around on two legs? Lots of people do
it all the time. You know, toddlers can get the
hang of it without too much trouble, and we celebrate
it when it happens like that's a big deal, But
then shortly after that it's It really becomes just a
source of stress as the toddlers toddle along toward one
danger or another. But you're entirely dependent upon just two
(14:51):
points of contact with your surrounding environment. If you're talking
about a true bipedal form that is capable of moving
around round the area, and those two points of contact
with the environment are essentially the bottoms of the foot seats. That's,
of course, if everything is actually going as you wanted
it to. If things have gone poorly, you might actually
(15:12):
have lots of different points of contact with the ground.
But that's because you went horizontal after something went wrong.
So you've got your robot. It's got two legs, the
two points of contact with the ground or the bottom
of the feet likely the legs and the feet, and
your robot in general has a number of degrees of freedom.
So degrees of freedom are joints that allow movement along
(15:34):
at least one axis. The point of contact with the
ground could actually be considered a passive degree of freedom
in itself. And you're also relying on friction to allow
your robot to stand up, to maintain balance, and to
get anywhere. If the robot's feet were frictionless, then it
wouldn't be able to stay up right at all, let
(15:54):
alone walk. And you've got all this weight above the
legs that you have to worry about. So have you
ever balanced something like, say, a baseball bat, on the
palm of your hand. If you do that with a
baseball bat and you're using the narrow part the handle
in of the words of the bat and you're balancing
that on your palm and the thick part of the
(16:15):
bat is up in the air, you know that little
motions can create big results. Right A small movement at
the base can cause the top to really sway, and
that requires you to make larger corrections down at the
base in order to keep everything in balance. Well, that's
kind of what it's like to try and figure out
how to make a bipedal robot walk while maintaining its balance.
(16:37):
You've got a inertia to deal with, and that really
affects balance. How do you make sure the mass above
the legs doesn't throw everything off kilter whenever it starts
moving or when it stops moving. How do you correct
for that so that your robot doesn't just tumble over?
How do you keep your bot upright? One early discussion
(16:57):
that became a core component in the pursuit of bipedal
robots revolved around a concept that came to be known
as the zero moment point or ZMP. So a pair
of smarty pants is from Russia named Beyomir Vukabratovich and
Davor Jurichic first described this back in the nineteen sixties. Now,
(17:21):
the actual term zero moment point would be coined a
little bit later, but it was used to describe what
they were talking about, and the whole concept revolves around
a moment in which the net reaction forces between a
bipedal mechanism's feet and the ground are essentially zero and
there's no movement along the horizontal plane. So that means,
(17:45):
like forward momentum and gravity they've kind of canceled each
other out. You've hit this zero moment where you don't
have to worry about the robot tipping forward and falling over.
It is you could think of as a moment of stability,
and the robot, assuming no external forces are acting upon it,
will remain upright, assuming that ZMP is maintained. Now, this
(18:09):
discussion really drives home how stability can be a huge challenge.
As robots move, they must deal with inertia and you
have to know the math to achieve dynamic stability if
your robot is to remain upright, whether it's walking, running, jumping,
or whatever. And in fact, it gets increasingly more difficult
as you go down those tasks. Walking is hard, running
(18:33):
is really hard, and jumping is really really hard. Like
the jump part might be easy, the landing and staying
upright that's the hard part. Otherwise your robot's gonna topple over.
And while watching a robot take a tumble might be
a great YouTube video. In practice, in the real world,
you obviously don't want your robots to be falling over.
You want your robots to be stable and capable of
(18:55):
moving around environments without causing damage or potentially injury to people,
to be able to maneuver. It's tough and robots are expensive.
You don't want them falling over. Then you're thinking, well,
that's twenty thousand dollars to get this thing back on
its feet. Again, that's not a great way to reach
progress either. So to plan the motion for a robot,
(19:18):
you need to be able to calculate the zero moment
point or ZMP. You need to figure out which joints
the robot is going to have to engage in order
to achieve stability under its various operating conditions, and those
conditions could include things outside of your strict control. It's
one thing to calculate how the robot can achieve stability
(19:38):
when it's walking across a level floor that has a
firm surface, But what about a floor that surface isn't
as firm. Maybe it's a little squishy, you know, maybe
it's a nineteen seventies shag carpet or something, or what
if the floor isn't level, what if the terrain is
actually uneven, you know, kind of like a typical sidewalk
in the city of Atlanta. How does a robot compensate
(20:01):
for all this, remain stable and keep itself from pitching over?
This is a non trivial challenge, and it takes a
lot of work to get to a point where robots
are sophisticated enough to achieve stability. Engineers have had to
take a lot into consideration. Would more degrees of freedom
help or does that actually overcomplicate matters? I mean, there's
(20:23):
no reason why we should be constrained to the same
degrees of freedom that a person has. Right Like, we
could think, oh, let's mimic the way humans work and
make sure that the ankles and the knees and the
hips all have the same points of articulation. Or we
could say, well, there's no reason why we couldn't have
more or fewer joints if it makes the operations work better.
(20:47):
So that's something to think about, you know. And then
walking would require shifting stability so that the robots can
maintain itself with just one foot in contact with the ground,
right like, suddenly points of contact have halved. If you're running,
it's even harder because with running, at least the definition
of running that roboticists use, there's a moment that may
(21:10):
only last just for a split second, but there's a
moment in which both of the robots feet are not
touching the ground. So how do you achieve stability when
your point of contact is continually interrupted? Jumping is even harder, right,
because you're really leaving the ground then, and how do
you ensure that landing you do so in a way
(21:32):
that you maintain stability. So that ZMP was a huge deal.
It still is a huge deal. Not all robot locomotion
is centered around ZMP calculations, by the way, but a
lot of it still is. So a lot of the
work in bipedal robots, particularly in the seventies and eighties,
involved calculating ZMP and keeping the robot within that zone
(21:52):
while engineering movements. So calculating ZMP is one thing. Building
a robot that can balance is another. Even building robot
that's capable of static stability is no simple task. Static
stability just means the robot is able to stand still
and not fall over, and believe it or not, that's
easier said than done. In nineteen seventy four, labs at
(22:14):
the School of Science and Engineering at Waseda University in
Japan started a project with the goal of building a
stable bipedal robot. This project became known as Weibot wabot
that stands for Waseda Robot. The labs created a new
focus group within the laboratory system at the university. It's
(22:36):
called the Bioengineering Group, and their first effort, the Waybot one,
wasn't exactly something that you would mistake as a human being.
It was not like a smooth humanoid robot. You would
never look at it and think it was anything other
than a robot. In fact, it looked kind of like
a giant erector set. It was very much a big, blocky,
(22:57):
metallic humanoid robot. So it had a limb control system,
obviously very important if you're going to have a walking robot.
It also had a primitive vision system. It even was
able to converse in Japanese to a certain degree. They
said that it had the in electrical capacity of like
a one and a half year old Keep in mind
this is the early nineteen seventies. It was tethered to
(23:20):
computer systems and power systems. There was not yet a
point where we had reached a miniaturization where you could
have all that computing power, not to mention electrical power
on board the robot itself. If you had done that,
the robot would either need to be huge or would
be carrying the biggest backpack you've ever seen in order
to have all the computational and electrical power necessary to
(23:41):
operate this thing. So, yeah, it was tethered. The Waybot
was the first humanoid robot to achieve static stability, and
later it would be the first digitally controlled anthropomorphic robot
at all that was able to achieve dynamic stability, and
statically stable means that the robot can remain balanced while
(24:03):
standing still. Dynamically balanced refers to robots that are maintaining
that stability while they are still in motion. Going from
one to the other and back again is actually really hard.
Like achieving one is hard, achieving the other one is
also hard. Going back and forth between still and moving
and maintaining balance is even harder. You know, remaining dynamically
(24:25):
stable is often easier than going from dynamic to static,
assuming that you know motions are smooth, and fast enough
to counter the forces acting on the robot, because when
you think about it, walking is really a series of falls.
It's like you're falling and you're catching yourself over and
(24:45):
over again. You move forward when you walk, not just
by moving your legs. You know, you kind of lean
forward into the walk as well, and for a moment
it's as if you're falling forward and that you would
face plant into the ground ahead of you. But of
course you've moved your leg to catch yourself, and this
process repeats, so you're propelling yourself forward by constantly almost
(25:08):
falling but catching yourself over and over and over again.
Now we do this without really thinking about it, like
once we learn how to walk, we don't have to
focus on this. This is this is just how we
do it. Robots, however, have to calculate this stuff in
order to do it properly, in order to catch themselves
with just the right amount of force to keep things
moving and not to use too much or too little
(25:30):
force and then risk taking a fall. Tons of work
in robotics continued around the world towards this goal of
creating bipedal humanoid robots, but nearly all the articles I
read for this episode. Make a big jump from the
nineteen seventies to the mid nineteen nineties. That's when Honda
unveiled a robot that they called the humanoid P two.
(25:53):
So Honda had actually been developing humanoid robots for about
a decade before showing off the P two. A series
that was kind of the in secret R and D
works that was called the E series of robots, and
this was in the nineteen eighties. These E robots were primitive,
but they demonstrate the capability to walk on level ground
(26:15):
under very controlled circumstances. The first E robot, for example,
was able to walk at an extremely deliberate pace because
it was said that each step took about five seconds
to complete. I challenge you to try walking that way,
even if it's just a couple of feet, count to
five slowly per step. That is slow, but it illustrates
(26:39):
how challenging it was to design a robot capable of
walking in a biped away. Now, the first E series
robot was built in nineteen eighty six. It essentially looked
like a pair of legs attached to robotic hips and
that's it like. It didn't have any top half. There
was no torso or arms or anything like that. The
B two wouldn't debut until nineteen ninety six. Honda had
(27:00):
a decade of work developing humanoid robotics before they revealed
to the public what they had been up to. And
like I said, E zero, not very humanoid. It is
like a free, free walking pelvis, robotic pelvis. The later
robots in the E series started to look I don't know,
in my opinion, even more strange, Like one of them
(27:22):
looks kind of like a microwave oven that has legs.
One of them looks sort of like the front end
of a fancy car, like a Rolls Royce, but with legs.
You would never call any of them human in shape,
but they were gradually evolving toward that. The P two
in nineteen ninety six looked much more human in that
(27:44):
it had legs, and it had arms and a torso
and a head. Now this head was rectangular, and it
was it was wider than it was tall. It was
not human looking at all, but it demonstrated that Honda
had been hard at work tackling this BikeE Hetle challenge,
and it would serve as a foundation for a much
more famous Honda robot just a few years later, a
(28:07):
robot that would debut in two thousand. It was called Asmo.
Now I have a fun connection with Asimo, and I
will talk about that, but first let's take another quick break.
(28:29):
All right, So the year was two thousand and seven.
Asimo had been a thing for the better part of
the two thousands, like it debuted in early two thousand
and now it's two thousand and seven. I had just
been hired by a company called HowStuffWorks dot com, and
one of my first assignments was to rewrite and to
(28:49):
update the article on how Asimo works. So I researched
the project, including the lofty goal that the engineers had
set to have a robot that couldn't just walk but
could also run. So it would be a robot that,
at little points would have both feet leave the ground
just for a moment, but still be able to maintain
(29:10):
its balance. Now that was a huge accomplishment. Even if
you were to watch videos and it kind of looks
like Asmo's doing a little hopping dance you might do
if you were in need of getting to a restroom,
Like I think of it as oh, it's doing the
peepee dance. But you know, watch Asmo running. It's cute,
it's weird, it has this odd sort of tone to it,
(29:30):
I would say, But it was phenomenal because yes, for
a split second, both feet are off the ground, and
yet when the opposite foot makes contact with the floor,
the robot would maintain its balance and be able to
continue running. Asimo looks a lot more human than P
two did. In fact, it looks like sort of a
(29:51):
diminutive astronaut in a spacesuit. I actually got to meet
Asimo once when I was at Disneyland in California, because
there was demonstration of Asimo that was a real blast.
I watched this presentation that they gave, and then I
mentioned to a cast member that I had written an
article about how Asima worked, and they brought me aside
and I got to meet the robot, which was a
really fun moment for me. That was kind of cool. Anyway,
(30:15):
Asimo would go on to establish a lot of firsts
in the bipedal humanoid robot space. Not only was it
the first one to run, it could also climb and
descend stairs, at least eventually it could. One early demonstration
didn't go so well, and Asimo tripped, but it later
demonstrated those capabilities, and these were things that would be
(30:38):
built upon in future robotic projects, both at Honda and elsewhere.
I should also mention that Asimo was largely a programmed robot,
in that it would maneuver around and interact with an
environment that had been carefully mapped out for the robot.
So it's not like it was spontaneous. It wasn't coming
into a brand new environment finding its way or around
(31:00):
picking things up that kind of stuff. It was following
a very specific set of instructions and it knew where
everything was supposed to be and where it was supposed
to go. So it was not like an autonomous robot.
But that really wasn't what the project was about. It
wasn't about autonomy. It was about robotic locomotion and interacting
(31:21):
in human spaces. So you have to keep in mind
that this whole approach to robotics is multidisciplinary in nature.
It requires lots of different work in varying fields, some
of which aren't even in technology. I'll talk more about
that later. In twenty fifteen, the Defense Advanced Research Projects
(31:44):
Agency here in the United States aka DARPA held a
robotics challenge. So DARPA is known for contracting with various
companies and research facilities to develop bleeding edge technologies potentially
useful for the purposes of national defense. They're not always
couched in those specific terms, but that's the directive of DARPA.
(32:05):
DARPA has played a part in everything from the development
of the Internet to the early days of driverlest car technology.
Well in twenty fifteen, they had a lofty goal set
for robotics teams and it was all inspired by a
terrible disaster. So in twenty eleven, an earthquake and tsunami
damaged the Fukushima Daichi Nuclear Power Plant in Japan. The
(32:28):
power plants backup systems were damaged, and this led to
a situation in which the plant's reactors began to overheat
because there was no power that could be used to
operate the cooling system in order to keep everything under control,
and ultimately this overheating led to a containment failure and
radioactive material was released into the environment. It was the
(32:49):
worst nuclear disaster since Chernobyl. Cleaning up after the disaster
was a really dangerous job. Response workers would be subjected
to potentially dangerous levels of radiation or extended periods of time.
DARPA's challenge was to give robotics teams a set of
tasks that a robot would have to complete with minimal
direction and intervention from the teams. The idea being let's
(33:12):
work toward a technological solution where we could develop robots
that could step in into situations that were like Fukushima
and take the place of humans so that human beings
don't put their lives at risk to do this kind
of stuff. The robots, since they have no lives, they
could go and do it and we wouldn't be putting
any human life in jeopardy. That was the concept. But
(33:34):
to do this, these robots would have to do very
human like things, and they have to maneuver in a
very human like world because it was designed by humans, right,
so no big surprise there. So the robots would have
to do stuff like get in, to get out of
and operate a vehicle, to be able to open doors
and step through doorways, to pick up a tool and
(33:56):
to use it properly. And teams were given some but
not all, the information that they would actually need in
order to complete the various tasks that DARPA had laid out,
and the reason why they weren't given everything is because
the whole concept requires teams to build a robot that
could accomplish goals in a world that is unpredictable and
uncontrolled for the most part. In a real emergency, there
(34:19):
may be no way to account for all the variables,
and depending on the nature of the emergency, a team
might not be able to maintain a direct connection with
their robot, so the robot would need to be able
to handle some of this autonomously. Now, most of the
challenges were pretty darn straightforward, and they would have been
trivial for most people to be able to complete if
(34:41):
they were given the assignment. You know, if you ask
your typical human to get into a vehicle kind of
like a golf cart and to drive to a specific location,
to then get out of that vehicle, to open a door,
go through the door, pick up a power drill, drill
a hole in a wall, climb some stairs, and navigate
some rubble, and you'd likely see a lot of people
(35:01):
succeed at this. It's again a pretty simple set of
tasks for most people, that's not a tough gig, but
for robots, it's a totally different story. Now. There are
compilations of videos of the various teams participating in this
competition that show just how challenging it really was. There
are videos of robots that, upon trying to just walk
through a doorway, fall over. One robot completed the list
(35:27):
of tasks, turned wave to the crowd, then fell over
because that balance thing is really hard, y'all. Creating humanoid
robots that can interact with the technology that was made
for human beings requires a ton of consideration and cross
disciplinary work. For one thing, sometimes it requires roboticists to ask, hey,
(35:48):
why did we make this thing work in this specific way.
It's almost like you have to go through reverse engineering
the world around you in order to understand why things
are the way they are. That doesn't always end with
an answer that makes much sense. By the way, sometimes
we're like, wow, we should really change this because this
is not the best way to do it, but at
that point it's the established way to do it. Ultimately,
(36:10):
a team from the Republic of Korea won this competition.
The winning robot was named DRC dash Qubo Hubo. It
completed the series of tasks in just under forty five minutes,
and the team took home a two million dollar cash prize. Now, no, lie,
two million dollars, that's a lot of money, But I
(36:31):
would be willing to bet that the two million dollars
doesn't remotely cover the costs of all the research, development,
and production of the robot itself. I bet if you
were to add up all the expenses of making this robot,
it would be more than two million dollars. But the
money wasn't really the full goal of this thing. Like
that was an award, but you weren't doing it to
(36:53):
win the money. It's this challenge trying to figure out
a way to achieve this really tough goal set out
by the nature of the challenge itself. That's the real
call the engineers out there, know what I'm talking about.
Like that thrill of tackling a problem and figuring out
a solution. That's really what drives a lot of engineers. Now,
(37:13):
if anything, the challenge illustrated just how hard it is
to build a humanoid robot that can function properly. Now,
the benefits are pretty clear. You know, this kind of
robot could potentially step in during situations like the Fukushima
disaster scenarios in which a human would be put into
danger and the robot, due to its design, could interface
with systems that had been built for humans. That's understandably
(37:35):
a worthy goal. It's just a very challenging one. And
it gets harder when we start to bring artificial intelligence
into this because we've been mostly focused on things like locomotion.
But let's talk about AI. Now. I've told this story before,
but I went to a panel and about robotics. It
was at south By Southwest. This was several years ago,
and at that panel, the presenters were talking about how
(37:57):
challenging it is to teach robots how to do things,
like not program the robots to do it, but to
teach them how to learn in an environment and then
replicate things that they have learned. Like even when you
build models in which the robots are able to observe
and then attempt to replicate actions, stuff can go wrong.
Robots have the same limitations as other examples of machine learning. So,
(38:22):
for example, I have often used an example saying, like
teaching a computer to do something like recognize that an
image represents a specific object like a coffee mug is hard.
Not all coffee mugs are alike, right, Some come in
different sizes or shapes or colors. Some might have handles.
Those handles could be shaped one way versus another, some
(38:42):
might be you know, the pictures of them might be
in under different lighting conditions, or they might be paired
with other stuff that's of a similar size or shape
to the coffee mug. All of these elements represent challenging
variables to machines that are being trained to recognize images.
You know, the machines do not inherently understand what makes
(39:03):
a coffee mug a coffee mug. That's what you are
teaching them. And you know, you can teach a human
being what a coffee mug is and they pretty much
get it pretty darn quickly, even to the point where
they can recognize other coffee mugs that don't look exactly
like the initial one. But for computers it requires a
lot more training. You train your model, then you tweak
all the settings so that you can improve your results,
(39:25):
you know, cut down on the false positives and fix
all the mistakes, and you train it again, and you're
potentially using millions of images in order to do this. Now,
consider the humble door, now a door is a pretty
darn simple thing to operate for most people, but there
are lots of different ways that a door could potentially
operate right Like, the door might have a knob that
(39:47):
you have to twist. It might have a bar that
you have to press, or a handle that you have
to pull. So when you encounter a door, chances are
you have a decent idea of how it works, but
you might not know which way it opens it for
that can give you a little bit of a pause.
One of my favorite of Gary Larson's Far Side cartoons
a classic cartoon strip from the eighties. It shows a
(40:09):
young boy pushing as hard as he possibly can on
a door, and just above his hand on the door
is a label that reads pull, and next to the
doorway is a sign that reads Midvale School for the Gifted.
I feel this cartoon in my soul some days. Well.
Robots are kind of like that student. Even a robot
that's been trained to open doors might need to pause
(40:30):
and have a digital think about things before giving it
an attempt. So the south By Southwest panelist was telling
the story of such a robot, and this robot sat
outside a door in a hallway. I think It was
an electrical engineering department at the university that this panelist
worked at, and it was sat there for several days
just contemplating the door, and it was really irritating folks
(40:53):
who worked there because they had to walk around this
thing in order to get through the hallway, and if
they passed in front of it, it irritated the row
bodicists because it could actually disrupt the process and set
things back even further. But the robot was just trying
to figure out how it should proceed in order to
try and open the door. And when you think about
how a robot could potentially be powerful enough to cause
(41:14):
damage to the environment it's in if it attempts to
do something incorrectly, then you start to understand why taking
time might actually be a necessity. It might be an
important thing to build into robots. It seems ridiculous to
just stare at a door for days on end before
even attempting to open it, but if you could potentially
(41:34):
rip the handle off the door or damage the door
in some way, well, taking time might be a needed precaution.
And that's not even getting into the challenge of having
robots that operate within an environment in which they're also
human beings. Obviously, you have to take a lot into
consideration in those kinds of situations where robots and human
(41:55):
beings are going to be working within the same environment.
Like in most industrial uses for robots, the robots are
very much separated from all the people, like there are
multiple safety considerations put in place to keep the robots
and people away from each other because the potential for
catastrophe is way too high. If you're working too close
(42:16):
to a robot, like it's a robot that welds stuff
or whatever. Well, you know you don't want to get
in the way of a welding, right, That would be awful,
It would potentially be deadly. So creating robots that are
capable of interacting among human beings it comes with its
own series of challenges you have to overcome. So not
only must the robot be able to move around without
(42:37):
falling over onto somebody, it needs to be able to
do this in a way that doesn't cause anxiety or
fear or other negative reactions among the human beings. Really
weird thing is that sometimes a robot can behave a
little too human and that can end up being almost
as bad as if it's not acting human enough. You've
got to find a balance, Like there are expectations that
humans have when it comes to interacting with the robots.
(43:00):
Of the robots behave too far outside that set of expectations,
it can cause issues. So robotics has become a truly
multidisciplinary endeavor. Making a bipedal humanoid robot capable of integrating
with humans the way the Tesla Optimist robot is supposed
to do that requires lots of work in disciplines that
(43:21):
go well outside of technology. We're talking about stuff like psychology,
and I think every time we see a remarkable achievement
in the robotics space, we're also reminded how far we
still have to go and how hard this really is.
So will Tesla's Optimist robots deliver upon all the promises
that Elon Musk often quotes at these events. Maybe I'm skeptical,
(43:46):
largely because Elon Musk has proven to make some rather
ambitious claims the past that have failed to manifest as described.
He's kind of the boy who called fully autonomous driving.
And largely I also have doubts because I have an
inkling as to how hard it's going to be to
make a bipedal general purpose robot that's at least as
(44:06):
good as, and hopefully better than a human being at
doing your typical tasks. If the robot is worse at
doing those tasks, then it's a waste of time and
money to use the robot. Just hire somebody else to
do it, it makes more sense. So like these are
these are really high barriers that you have to you
have to get over, and I don't think we're going
(44:28):
to get over them very quickly. I think it's going
to take years and years more work. But it is
a heck of a goal to aim for. I don't
want to shame anyone for taking aim at achieving this
really difficult task because it drives innovation. I think that's important.
So I don't want to dismiss anyone who's working toward
(44:49):
building bipedal humanoid general purpose robots that have a level
of AI that allow them to operate autonomously within a
human environment. I think that that is a phenomen I
just think it's also one that's going to require many
more years of work for it to be a viable project. Right, Like,
I guess I could see a disappointing version becoming a
(45:13):
reality within a couple of years. But that's really falling
fall short of the promise, and I would much rather
see more work being done to improve the technology than
for a premature release of some humanoid robot that just
doesn't do anything well enough to justify its existence. That
would really take a lot of wind out of the sales.
(45:36):
I think, all right, that's it for this episode of
tech Stuff. I hope you're all well, and I'll talk
to you again really soon. Tech Stuff is an iHeartRadio production.
For more podcasts from iHeartRadio, visit the iHeartRadio app, Apple Podcasts,
(45:58):
or wherever you listen to your favorite shows.
TechStuff News
Hosts And Creators
Oz Woloshyn
Karah Preiss
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.