First posted Feb 2005
OVERVIEW
It is an interesting
paradox in robotics that analogies with human body parts are both
immediately useful and at the same time extremely self-limiting.
Unlike human
body parts, robot parts can be mixed and matched. As one example,
the same operating system that enables a small robot to clean
sludge out of a pipe might also be used for a Hollywood entertainment
robot. A robot company might have to get established in business
by first making money in a conventional waste application before
broadening to the Hollywood variety.
The
future is in plastics, young graduate. Also unlike
humans, robot dimensions have a high degree of plasticity (or
malleability). Through re-engineering, they can be shrunk, expanded
in size, or chopped up into multi bot systems. In addition to
serving as self-contained systems, they can also become embedded
inside existing machines and also inside plants, animals, minerals,
and things.
If we change
metaphors from human body parts to a poker hand, lets look at
the cards that we are currently being dealt when it comes to the
state of the art in robotics modules:
BRAINS:
The Neal
Scanlan Studio Performance Animation Controller (PAC)
used by Hollywood |
No need to
feel threatened quite yet, since robot intelligence currently
rivals insects. Sometimes it rivals two month old human infants
as well. (Except for when a robot is hooked to a supercomputer,
which I will get to later). It may be another ten to fifteen years
before the average autonomous robot functions on the level of
dogs. Add on another five to ten years after that before robots
rival most humans. For folks who can’t wait, there are four
strategies worth mentioning. They are remote presence, layering,
external database management, and (you guessed it), augmenting
external processing power via a supercomputer.
Remote
presence: In this approach, human and robot brains
work together in partnership. Imagine, as one example, that we
have a human standing in front of his computer screen. He is connected
by Internet to see the camera vision of a robot located far away.
He wears a full body exoskeleton that extends over his hands,
arms, torso, and legs. It has a cord that attaches like a mouse
to his computer. As he moves his hands, he sees the robot move
its humanoid arms and hands in the exact same way. As he guides
a robot hand to grasp a pole, the haptics (touch) system
allows him to feel resistance in his own hand as the robot grasps
an actual object. SARCOS robotics has an excellent demo video
web page
of humans with full suit and hand exoskeletons manipulating robots.
The Walt Disney
Imagineering Company coined the term "animatronics"
in the 1960's to connote "robotic puppeteering" for
Disney theme park creatures and other entertainment applications,
so the underlying concept has been around for a while. The Scanlan
PAC depicted above is an example of a system that enables human
hand and finger movements to bring fantasy creatures to life.
Where the
future really becomes now in the remote presence
concept is where an artificial intelligence program can remember
all the moves made by a human and even fill in voids in instructions
to give the robot various measures of autonomy. As an example,
if a human directs a robot to climb some stairs, the human does
not have to guide each step. In addition, the robot is able to
comprehend when it has reached the top of the stairs to go back
to a normal walking mode on its own.
At the 11-14
Oct 2004 Carnegie Mellon Robotics Institute 25th Anniversary Event,
I listened to neuroscientist Dr. Mitsuo Kawato, Director of ATR
CNS Laboratories in Japan, discuss a collaborative R&D
project near Osaka using a SARCOS robot. He has called on the
Japanese government to commit $446 million a year for the next
three decades for the "Atom
Project." He wants to build a robot with the mental,
physical, and emotional capabilities of a five year old child,
and mimic inside the robot all the different functional areas
of the human brain. Already the SARCOS
robot has the autonomous ability to play a drum and bounce
a ball on a racket. Dr. Kawato showed us a video featuring a Japanese
TV reporter having fun playing a scrabble game against the robot
with Georgia Tech researcher Dr. Chris Atkeson standing between
the contestants.
One might
imagine agricultural applications where people who live in deserts
or economically depressed areas would be happy to offer their
services over the Internet from their homes to operate robots
in lush areas. Complex farm tasks might be broken into lots of
little tasks handled by an “Ag
Ant” or "coolie-labor" army of robots. This
is another way one could bring factory assembly line concepts
to the field. As individual robot behaviors become increasingly
shaped with artificial intelligence programs, a human handler
could supervise more robots simultaneously on a split screen on
his computer.
Remote presence
can not only mean one person controlling more robots, but also
the other way around where many people guide one robot from different
locations. Dr. Robin Murphy, Director of Center for Robot-Assisted
Search and Rescue in Tampa, FL, made this point at her Oct 2004
talk at CMU. As search and rescue robots were going through the
World Trade Center rubble following the 9-11 disaster, the grayed-out
appearance of dust-covered objects made it hard to identify what
they were and undermined depth perception. The more people with
expertise in different areas watching the robot camera pictures,
the better the interpretative process. In addition, when search
and rescue robots are racing against time to save victims, overlapping
viewers and operators eliminate robot downtime. According to Dr.
Murphy, about 20% of collapsed building survivors are entombed,
and their life expectancy rapidly diminishes after 48 hours. Also,
in addition to looking for survivors, emergency personnel have
an urgent need to assess the probability of further collapse,
trace the origins of any smoke, and conduct other immediate surveys
and assessments of the site. People in their homes or offices
in far away cities can just as easily be watching the cameras
and take over the controls as someone on site. Lastly, getting
back to the aforementioned "Ag Ant" concept, it may
be better to have lots of little robots swarming around, each
carrying their own cameras to view the same object from different
angles, and each being monitored by many different people in different
places, than to have everything concentrated on one robot.
For remote
presence applications by wireless or over the Internet, the delay
in round trip signal time (or its latency) can become a big issue
depending on the distances and switching units involved. An MIT
study in which humans controlled robots moving nuclear materials
determined that when latency increases over half a second, a person
is unable to keep track of the actions commanded. (Brooks,
p. 133). The half second delay issue usually occurs when someone
is trying to operate over the Internet from the other side of
our planet.
Currently
RoboDynamics
of Los Angeles appears to be leading the charge to produce a lower-priced
remote presence robot platform. Its MILO,
the personal robotic assistant, is priced at $2,999. MILO can
serve as a remote mobile sentry for home or industrial use. It
consists of a camera and microphone mounted several feet above
a mobile circular base. On each side of the camera eye are panels
that hold various types of blinking lights that have been programmed
to show twirl or star patterns to indicate certain moods. The
operator uses a joystick to control the robot's movement and has
a panel menu on his computer to select certain blinking light
moods, make certain sounds, or click a digital snapshot of the
current screen view. As part of its strategy to promote early
adoption and innovative feedback, RoboDynamics offers a $500 discount
to tinkerers who qualify for its early adoption
program.
On the specialty
high-end arena, Pittsburgh-based Mobot
has produced socially interactive robots with remote potential
for museums and for a classroom "field trip" inside
an aviary.
I personally
think the really big mass market commercial applications will
require certain features that have not yet come down far enough
in price or reduced "intellectual load" (difficulty
of operation). They will likely include remote hand manipulation,
artificial intelligence behavioral shaping, a high degree of
autonomous learning and sensing, extreme ease of control,
easy navigation
systems, a modularized ability to fit on a wide variety of exotic
propulsion units, and two way video-conferencing. The latter
may
include snake robots that reach disaster victims amidst rubble
or spider bots who work beside oil workers on rigs.
On the commercial
hand manipulation side, surgeons are currently using remote presence
robots created by Intuitive
Surgical Corp (Nasdaq: ISRG) for serious procedures. The surgeons
insert their hands into loops that control robotic
manipulators elsewhere in a room. A brain surgeon can damp
down his precision movements by gearing the robot’s movements
in ratios such as one to five relative to his own. The robot system
costs about $1
million and can just as easily be located 100 km away as in
the same room. (Brooks,
p. 224).
In the mobile
video-conferencing arena,
In Touch Health has developed the RP-6, informally dubbed
“Robo Doc” by some journalists and hospitals. It allows
a doctor at home or in his office to control a robot in a hospital
over the Internet. Currently the robots are being used in emergency
rooms and intensive care facilities, and their use is growing.
A doctor can navigate an RP-6 down corridors into patient rooms
or onto elevators using a joystick. Sensors and safety software
help to avoid inadvertent collisions. The latency factor between
commands and robot response over the Internet is about 200 milliseconds.
In Touch Health's research suggests that this latency has been
very acceptable, and does not become an issue until it approaches
400 milliseconds. (This is consistent with the .5 millisecond
maximum latency factor determined by the MIT study mentioned elsewhere
in this article). Interestingly enough, In Touch is finding doctor
acceptance without including such Robonaut-like features as hand-manipulators.
The company is focusing on making the system as user-friendly
as cell phones to speed its acceptance within their specialty
niche.
Another interesting
variation of the remote presence concept in the propulsion and
ease of control arena is the Intelligent
Assist Device (IAD). In this case, the robot is actually on
the human himself. As the human initiates a movement with an arm
or leg, the robot augments it. As an example, Hiroshi Kobayashi,
one of Japan's leading robot scientists, has designed a muscle
suit to help disabled people move about. An exoskeleton can also
be used by industrial workers and military
personnel to help lift heavy objects.
Apart from
these kinds of specialty niches, the most obvious use for current
remote presence technology at current prices is security/surveillance.
The problem here from a business viewpoint is that the robot competes
with the static "sensor suite" concept. It is still
much cheaper to rig lots of cameras, microphones, and other sensors
in one location than to buy, operate, and maintain a mobile robot.
But as processing power steadily increases and component costs
steadily decline, this some day will change.
Layered
intelligence: The human body actually consists of
several “brains” in different locations that interface
with each other. If all human motion had to be directed by “central
planning” from the skull area, we would not be able to walk,
digest food, or keep our hearts beating. Thanks to the autonomic
nervous system, when we touch a hot object, out hands pull away
faster than nerve impulses travel to the brain and back.
Subsumption
architecture refers to layered and distributed intelligence
in a robot. This is analogous to the autonomic nervous system.
One example is the robot Genghis, now in the Smithsonian Air &
Space Museum, built by Dr. Rodney Brooks and Colin Angle in the
late 1980's. Intelligence modules at lower levels allow the legs
to immediately react to their environment in certain ways that
are not controlled by the higher "brain."
Another example
involves robotic "cockroaches" being built at Case Western
University. The robot designers are trying to replicate nature
as much as their man-made materials will permit. Although cockroaches
in nature appear to move very quickly over uneven surfaces, slow
motion studies reveal that their myriad feet actually stumble
and bumble and feel their way along in fast motion, and are not
precisely coordinated by whatever a cockroach has for a main brain.
| 
Robo-Knee demonstration, photo
© Yobotics |
Last, but
not least, layered intelligence can also be found in robot prosthetics
(artificial limbs for humans). While attending the March 2004
robot show in Cambridge I saw a video of a girl with an amputated
leg. The “before” video showed her pivoting sideways
as she struggled to hobble up and down stairs with a long stiff
artificial leg. The “after” video showed her using
a robotic leg produced by start-up firm Yobotics
of Boston, MA. The robot leg has an autonomous sensing capability
to flex at the knee. The girl was able to walk up and down stairs
looking fairly close to normal.
I was part
of a large audience that watched this video. It drew a strong
emotional response that led to spontaneous applause.
External
Database Management: This is another important "brain"
function for robots. ActivMedia
CEO Jeanne Dietsch described an example of a data-gathering bot
in an interview.
“Hewlett-Packard uses our PatrolBot in their data centers.
It's got temperature sensors. It drives around several times a
day, and the data from these patrols is used to create 3-D models
of the heat in the facility. If there's a problem in the facility,
the computer or a human can send the robot out to check on it.”
According
to David Hyams, chief technology officer of Seattle-based robotics
systems integrator Coroware,
a robot can be the tip of the iceberg of a database management
and retrieval system. Whether a task involves picking grapes or
checking for metallurgical stress, the robot can sense and permanently
store and analyze everything that is relevant about the jobs that
it is doing. As an example, let us imagine a robot of the future
that grows grapes. The robot/data system can monitor how much
water and sunlight each plant is getting, and know exactly how
much water or artificial light to add to produce the desired quantity
and quality of grapes.
If
all else fails, connect to a supercomputer.
When earlier I compared average robot intelligence to the level
of insects, I was referring to what can normally be stored inside
a mid-sized autonomous robot. Sony recently gave its 23 inch tall
QRIO
humanoid robot a colossal"swelled head" by using
broadband wireless to connect it to 250 personal computers. As
another example, Dr. Rodney Brooks can boost his robot Cog
(short for “cognition”) into a mini-HAL in the MIT
research lab by hooking it up to dozens of computers racked and
stacked together. Many of them are on different operating systems,
but can still process data together.
250 personal
computers linked together takes us to where chip sets inside the
average robot will probably be in about five to ten years.
Currently,
the biggest constraint for these jury-rigged supercomputer systems
involves developing robotic software that can effectively use
all this capacity. In keeping with his “fast, cheap, and
out of control” (or “bottom up”) approach to
Artificial Intelligence, Dr. Brooks, his project team leader Brian
Scassellanti, and other staff members are teaching Cog how to
perform simple tasks in an order analogous to the way that a human
infant matures. One of Cog’s more recent projects has been
to learn how to play with a slinky toy and follow people around
the MIT lab with its camera eyes. According to Dr. Brooks, his
robot, “Knows how to sling a coil, but can not compare one
coil with another. To do that we would need to get to the intelligence
level of a three year old, and we are not there yet.”
NAVIGATION
To date, most
mobile industrial robots have required such navigational tools
as bar codes on walls, transmitters, or tape on floors.
For farming
applications that use heavy equipment, which are not constrained
by the size, weight, or cost of the navigation systems, agricultural
engineers at the University of Illinois have
already developed completely autonomous robotic tractors that
can systematically go down one row, turn, and then go down the
next row. A remaining issue is how to detect and avoid running
over a pipe or some other item left lying in the field.
Outdoor satellite
GPS systems have played a major role towards increasing autonomous
operation. So have newly developed indoor GPS systems. Arc
Second Navigation’s 3-D grid coordinate indoor laser
system enables mobile robots to approach aircraft or vehicles
in a hanger and know exactly where to place rivets, perform scans,
or do other useful work.
Still
in drag mode. For many robotic applications, the
navigation system is the core of the “robot.” With
certain types of heavy equipment, the propulsion systems and work
devices come ready made. This is why considerable robotic research
has been focused on aircraft, ground vehicle, or farm equipment
applications. At the March 2004 robot show one of the speakers
voiced concern that this has created a robotic development mind
set overly focused on “dragging things around.”
| 
The Stanford Team robot vehicle
finished first in the DARPA race
|
Getting
racy as well. A holy grail of navigation R&D
has been the annual DARPA Grand
Challenge. On Oct 9, 2005, the Stanford
Racing Team won a $2 million grand prize check. Their robotic
vehicle traveled the 132 mile course between Los Angeles and
Las Vegas with only robotic navigation. Four other vehicles made
it across the finish line as well out of 23 entrants. This was
a big improvement over the prior year, where none of the robo-vehicles
got very far. In 2004, Carnegie Mellon’s robotized Humvee,
traveled the greatest distance of seven miles. In the
2005 race, Carnegie Mellon supplied the second and third place
winners.
A bigger
potential prize than the $2 million or the prestige may be
the prospect
of future Army developmental contracts. A few years ago the U.S.
Congress mandated the Army to have as its goal one
robot for every third Army vehicle by the year 2015. So
far the big money robotic navigation contracts have gone
towards
cruise missiles and predator drones.
If
you could see her through my eyes… In terms
of building a robot system that tries to see in a way that is
similar to humans,
SEEGRID's NavSystem (a project of Dr. Hans Moravec mentioned
in Part
One) comes closer than almost any other system. It uses stereoscopic
vision from cameras to build a point by point 3-D map using ambient
light. At the Robotic Institute 25th Anniversary, Dr. Moravec
told me that some time in the not too distant future this should
allow robots to navigate their way around warehouses with reasonable
accuracy on ambient lighting alone using relatively cheap cameras.
Assistware,
founded by Dr. Dean Pomerleau and Dr. Todd Jochem, has developed
a robotic “drowsy driver” technology. It tracks driving
patterns and helps alert drivers if they make un-signaled lane
changes. According to various studies, driver fatigue is a major
cause of traffic fatalities. About one out of five drivers fall
asleep at the wheel.
| 
Evolution
vision software outlines Kermit the Frog |
In
a Flash:
Canesta Corp
in San Jose uses chips that track the travel time of bursts of
light off objects to build a 3-D image.
SICK has
developed lasers that create 3-D maps of underground mines.
This helps
miners monitor extraction progress and avoid
having one tunnel run into another. In my paper “Mining
and Robotics” I discuss advanced systems developed
by MD Robotics of Canada, which has partnered with Atlas Copco.
A different firm called Workhorse
Technologies, founded by Dr. William L. Whittaker of Carnegie
Mellon, has created an interesting application that maps flooded
or abandoned mines potentially suitable for reuse.
Getting
back on the right path: “Seeing” and
interpreting the environment are two different things. One test
of a robot’s navigational capability is whether it can reorient
itself when a human picks it up and moves (or "kidnaps")
it to a new spot. Evolution
Robotics offers its vSLAM
technology that enables to a robot to create a visual outline
of its environment. It performs a statistical calculations to
establish landmarks. Evolution can also overcome kidnapping with
its NorthStar
navigation system that enables a robot to determine its location
anywhere in a room by sensing two infrared spots projected on
a wall.
Evolution
Robotics has also made impressive advances in object recognition.
Its LaneHawk
system recognizes products on the bottom of shopping carts,
even when they are turned at odd angles or partially obstructed
by the legs of the grocery shopper. Computer processing power
has developed to the point that the vision system can recognize
a bag of potato chips or can of dog food turned at odd angles
without using bar codes. Some experts estimate that grocery stores
lose an average of $10 a lane a day by a failure to ring up "bottom-of-basket"
items, which amounts to about $1.8 million per store per year.
Both this object recognition technology and the vSLAM technology
are also currently used by Sony's AIBO in its "pet"
application.
SENSORS:
This topic
offers another paradox. On the one hand, humans have devised probes
that can sense all of the invisible (to humans) as well as visible
parts of the electromagnetic spectrum. Humans have developed mechanical
devices that can not only simulate their other senses (taste,
smell, feel, hearing), but can also exceed the sensing capabilities
of various exotic animals in nature. It is relatively easy to
mount myriad sensors on robots so that they can perform a wide
variety of potentially profitable tasks ranging from surveillance
to inspecting the structural integrity of various materials.
On the other
hand, it can be relatively hard for autonomous robots to make
sense of their sensor inputs, particularly to aid mobile robot
navigation and manipulation.
Computers
can detect people’s faces in a scene, but have problems
recognizing people from non frontal views or as they age. Dr.
Brooks points out, “The truth of the matter is that we have
no computer vision system that is at all good at recognizing that
something is a cup, or a comb, or a computer screen. Our computer
vision systems can do a few things with great skill, but still
after forty years of effort they are not good at the things we
humans and many animals do effortlessly. Because of the increase
in computer power over the last thirty years, we can no longer
blame a deficiency there on our poor computer algorithms. It is
clear that we must be missing something fundamental in the way
that vision in humans is organized, although almost no one will
admit that.” (Brooks,
page 90).
Dr. Brooks
points out that when humans try to program machine vision with
algorithms on a pixel level, it is hard to instruct a robot how
to cognitively outline a pen sitting on a desk from the desk itself.
He also talks about how the brain reconstructs a coherent field
of vision to overcome a blind spot that provides neural and blood
connections in the back of our eyeballs (Brooks,
p. 77). We need to not only be able to sense the images of things,
but must also be able to mentally model, filter, and reconstruct
what we are seeing so that we can recognize the most important
aspects from extraneous background.
I am aware
of studies in which human infants become disturbed if shown human
faces with third eyes. Also, no one needs to school most humans
on how to mentally outline and sexually respond to erotica. Obviously
quite a lot of outlining and reconstruction in the human brain
is innately programmed. On top of that, I would guess that our
brains must store hundreds of thousands, if not millions of “mental
movies” of 3-D rotating images that we have acquired from
interactive experience since childhood. We also have vast mental
libraries of symbolic connections, such as how a silhouette of
a tree also signifies a tree, or how the kinds of subtle clues
a detective might look for in a murder mystery can link back to
the “tree” concept. Anyone who has observed the megabytes
of memory consumed when they download video MP3 files can appreciate
how much processing power all of this must require. All of this
may help to explain the other paradox I mentioned earlier in Part
One that ditch digging utilizes vastly more processing power than
adding a column of numbers.
END
EFFECTORS AND MANIPULATORS
For certain business applications, developing a better end
effector may be more important than developing a better robot.
An end effector can be anything that does useful work, such as
the plow unit of a tractor, an x-ray device to check for metallurgical
stress, or a bulldozer blade that helps remove contaminated debris
at a nuclear site.
Many end effector
areas are undergoing their own steady evolutionary process where
quality is steadily increasing as prices are coming down. As one
example, according to Ralph Miller at General
Lasertronics, the power of Yttrium Aluminum Garnet (YAG) lasers
has increased more than ten fold over the last seven years, while
the size of laser heads connected by fiber optic cables has decreased
along with the costs per unit output. Among other things, these
lasers can be used to remove paint or to ablate rust and contaminants
off of various surfaces.
A good example
of the state of the art in remote presence hand manipulation is
NASA’s robonaut
B project. This is a portable device that can be mounted
on wheels for planetary surface mobility or on special legs
to attach to the surface of a satellite. The sensitivity of hand
manipulation is being enhanced through the use of micro components
and other nanotechnologies.
PROPULSION
Certain men once looked at birds and decided to invent human-built
flight. Certain men are now looking at other animals moving through
the environment in unusual ways (relative to man), and with the
aid of robotics, trying to do that too. Develop it and patent
it, and show how this platform can carry an end effector that
does useful work, and the world may beat a path to your door.
Inuktun
Services Ltd. in British Columbia and RedZone
Robotics in Pittsburgh are leaders in commercially successful
robot configurations that crawl through pipes. This concept gets
even more creative with snakebots
that can roll, climb, swim, and move in sinusoidal patterns, or
the "polymorphic"
snake or spider-like robots that can break themselves apart into
pieces and recombine into new forms. The latter would be particularly
effective to penetrate rubble to bring aid to disaster victims.
At universities
around the world, scientists are modeling robots off just about
any kind of animal you can think of, ranging from lizards and
birds to crabs and kangaroos. These robots serve as ongoing, interactive
"lab experiments" and "show me" tinker toys
regarding what scientists either know or still do not know about
how various animals move. I would like to touch upon three sample
areas that are profiled in the book Robo
Sapiens, Evolution of a New Species by Peter Menzel and
Faith D'Aluisio.
Quinn
and Ritzmann with their robo-roach
photo:
©
Robo Sapiens: Evolution of a New Species
..(Peter Menzel
and Faith D'Alusio/The MIT Press)
Cockroaches
can move fifty
times their body length in one second, which on a human scale
is the equivalent of 200 mph. Two leaders in cockroach
mobility research are Roger Quinn, a mechanical engineer, and
Roy Ritzmann, a biologist, at Case Western Reserve University.
They have spent many years refining a 16:1 scale robot replica
of a cockroach. As mentioned earlier, they watch slow motion videos
of cockroaches in action for clues about ways to replicate the
cockroach's decentralized nervous systems (subsumption architecture).
They have run into weight
and stress problems using steel springs and tubes to model the
stumble-bumble fast motion of cockroach feet. They hope to find
better answers in "plastic muscles" (or electro-active
polymer "artificial muscles" as described later -author)
which more closely replicate the properties of organic materials.
(RS,
pages 102-105)
Dr. Robert
J. Full at UC Berkeley discovered in 2002 that gecko feet are
covered with millions of tiny hairs called setae that allow them
to walk along walls and ceilings. Dr. Fuller has worked with Alan
DiPietro at iRobot, who developed a tiny $600 robot that uses
synthetic hairs to replicate gecko climbing behavior. (RS,
pages 91-93). UC Berkeley now holds a patent to a material that
works much better than most bandage adhesives for medical applications.
Certain fish
such as pike can accelerate underwater at a rate of eight to twelve
G’s, which is as fast as any NASA rocket. Scientists have
not figured this one out yet. However, you guessed it, Dr. John
Kumph at Draper Laboratories in Cambridge, MA has built a 150
kg robo-fish to try to get some answers. The US Navy is also interested.
One can not only envision faster torpedoes, but “fish”
that can conduct surveillance on enemy subs or minefields. (RS,
pages 108-109)
In his October
2004 talk at CMU, Dr. Full pointed out that biological inspiration
can be pushed too far. Some of man's most impressive engineering
achievements, such as wheels and gears for motion, do not have
a direct animal analog. Also, evolution is often a zig zag path
that leaves animals with structures that lack engineering sense,
such as the pelvic bones in whales. Lastly, 85% of world animals
are arthropods with an average length of 5 millimeters, obviously
creating a scale problem for humans.
The ultimate
aim is to engineer beyond animals by first learning the secrets
of their natural engineering and functionality. The latter can
include redundancy, which allows animals to continue functioning
when they lose an organ or limb, and self-assembly, which allows
animals to grow or heal themselves. Once they learn the secrets,
humans can scale up or scale down (that is, use nanotechnology
to produce nanorobots) and mix and match capabilities to serve
specific robotic purposes.
Reducing
costs and mechanical complexity: Dr. Mark Cutkosky
is one of many researchers at Stanford and SRI International who
develop electro-active plastic polymers that contract like muscles.
They are activated by embedded sensors and motors, and have no
moving parts. Like muscles, they can not only produce motion,
but act as shock absorbers. In the paper he coauthored "
Fast
and Robust: Hexapedal Robots via Shape Deposition Manufacturing
(SDM),” he argues that SDM techniques (which create artificial
skeletal structures) can be combined with artificial muscles to
propel at a rate of over four body lengths per second. Artificial
Muscle, Inc. was funded in March 2004 to capitalize on a market
estimated
at over four billion.
If legs made
of these polymers could be swapped for the wheels and treads,
we might see a substantial reduction in the number of parts and
subassemblies. We might also see a lowering in costs without a
significant degradation of capabilities. The use of polymers might
also enable more inherent fluid
motion with less need for nervous system direction, a topic
of great interest to the aforementioned cockroach researchers.
Walk,
jump, jog the humanoid way. Sony's QRIO is a leader
in independent stable bipedal motion control. Yes, there are people
at MIT's
leg lab and other American research institutions who have
impressive mobility projects, but you can not take anything away
from Sony's elegant execution.
Sony's QRIO
has four pressure sensors on the soles of each foot to balance
on uneven surfaces. It can cushion itself when tipped over, and
pick itself up whether it falls forward or on its back. It has
pinch detection sensors so that its limbs go limp when brushing
against a person to avoid injuries. It has compliant grasping
fingers that can throw a ball. It has precision-engineered joints
that are quiet and dependable, as demonstrated in videos.
It can even kick a small ball.
Safety
in numbers: Another novel approach to propulsion
is to exploit the design malleability of robots. Rather than try
to get a big robot to go where you want to go in one big piece,
one might consider getting there with the surviving remnants of
lots of little pieces, or little pieces that are carried as adjuncts
to the big piece like lifeboats. This dovetails with the. midget
submarine approach mentioned later, which is a variation of
the "swarm bot" concept.
COMMUNICATION
A good starting
question for this area is: “Who is communicating to whom?”
I would like to touch on three areas: human to robot, robot to
robot, and Internet to robot.
Human
to robot: The easiest way to interact with a robot
is to give voice commands. As mentioned earlier, Sony's AIBO can
respond to 75 simple voice commands.
The fundamental
problem with voice communication is essentially the same that
I described earlier with visual understanding. When we listen,
we associate, filter, and reconstruct what we hear just like what
we see. We draw upon mental libraries consisting of hundreds of
thousands, if not millions of thought models. Our speech draws
heavily upon visual metaphors or visual interpretations of the
environment. If a computer lacks our interactive visual understanding
of the world, how can it say back to us "I see what you mean?"
I am reminded
of Anne Sullivan's problems in connecting the world of abstract
thought to 12 year old deaf, blind, and mute Helen Keller in The
Miracle Worker.
Our ability
to understand abstract thoughts beyond simple nouns, simple verbs,
or simple noun verb combinations (for example “I eat”
or "you run") is called syntax. Humans have syntax,
other animals such as gorillas and chimpanzees do not. (Brooks,
pages 3-4). It may take fifteen to twenty years or more for the
average robot to have the internal capacity necessary to rival
human capabilities this area.
Currently,
most human instructions to robots consist of writing lines of
code. Evolution Robotics is developing software to help bring
programming into a format reminiscent of the menu-driven, point
and click Microsoft Windows environment. According to Coroware's
David Hyams, there are folks at Microsoft who are beginning to
wake up to the need for making Windows XP embedded more compatible
with robotics. He himself likes to install wireless communication
capabilities in all his robots so that he can flip open a hand-held
wireless LCD device any time he is around them and query their
programming.
Robot
to robot: The ability of robots to immediately update
each other by wireless or by some other means will create business
opportunities for risky, exploratory situations. As one example,
an Australian group argues that it is more cost effective to explore
undersea regions with schools
of midget robot submarines that swap information than to risk
larger subs with human crews.
Internet
to robot: Much of the current thinking in this area
involves linking robots to databases available over the Internet,
or creating robots that serve as mobile, Internet-connected personal
computer platforms for humans.
I think that
an especially exciting application will some day involve the ability
to download behaviors, skill sets, and professional analytical
capabilities over the Internet, either on a pay basis or as freeware.
As one example, a real estate developer may design standardized
rooms with transmitters embedded in walls to aid robot navigation.
As an inducement to buy or rent his units, he can develop downloadable
behavioral packages designed to show home robots how to go about
their cleaning chores within each unit. Since most household robots
will probably be fairly stupid over the next ten years, they will
need precise instructions regarding how to orient themselves within
each room and how to go about each task. . As another example
much further into the future, imagine that a couple wants to lead
a relatively self-sufficient in a wilderness area. Imagine that
they can download behavioral freeware that show their robots how
to perform such tasks as building a wood frame house, creating
and tending to gardens, and helping to home-school their children.
In regard to the latter, imagine a humanoid robot with a facial
screen similar to Robo Doc that can assume the facial appearances,
movements, gestures, and speech of some of the world's best teachers.
POWER
Power remains
a huge constraint for unplugged, autonomous robots. Batteries
are not much different in size and weight today than they were
ten years ago. When a robot designer tries to increase battery
power, he typically increases battery weight. More weight consumes
relatively more power for movement. There is obviously a rapid
rate of diminishing returns here.
Roboticists
are using a number of strategies to deal with the battery problem.
One very obvious approach is to design the lightest and smallest
robots possible. Another approach is to reduce electrical power
consumption and heat generation by redesigning robot chips and
other robot hardware, an area where VIA
Technologies has been a leader. A third approach is to make
robots autonomously rechargeable, either by carrying solar panels,
or by creating the ability to autonomously re dock at recharging
stations. The Roomba and Aibo both have this capability. A fourth
approach involves the most obvious of all -use an electric cord
or some kind of hybrid power system that produces electricity.
A
summary remark regarding modular frontiers: Some
areas of robotics are advancing at a rapid rate, while other areas
(such as power) are only plodding along. In many ways the field
seems disjointed. This is all the more reason for analyzing the
different pieces of the technology jigsaw puzzle in detail. I
address this and other issues from a business perspective in the
next section.
Link to......
Part
IV of series
Jump
back to......
Part
One.......Part
Two..
Jump
forward to......
Part
Five.......Part
Six
Disclaimer: This
report is for research/informational purposes only, and should
not be construed as a recommendation of any security. Information
contained herein has been compiled from sources believed to
be reliable. There is however, no guarantee of its accuracy
or completeness.
Bill Fox is VP/Investment Strategist, America
First Trust. Bill welcomes phone calls and email responses to
this article. His most current contact
information is at his web site: www.amfir.com.
Short URL for this web page: http://tinyurl.com/29xppdr
at the Battle of Cowpens, S. Carolina, 1781