RALPH HOLLIS is a robotics specialist. And he has bad news. Year by year, the
circuit boards, microchips and liquid crystal displays in everyday electronic
products get smaller and so have to be assembled with greater precision. But the
robots in today鈥檚 factories are no longer up to the job. They are too big and
too clumsy to handle the smallest components, he says. And as electronic devices
and the components they are built from get smaller, the problem will get
worse.
So Hollis has come up with an alternative. At the Robotics Institute, part of
Carnegie Mellon University in Pittsburgh, he is designing a new generation of
industrial robots no bigger than coffee mugs that float on cushions of air and
manoeuvre around an assembly line almost small enough to fit on a desk top. Each
robot will concentrate on one simple task, but will work in a team of up to 40
companions to assemble electronics goods with unheard-of precision. And because
electronics equipment has a habit of becoming outdated in a matter of weeks
rather than years, Hollis is designing his robots so that the tabletop factories
of the future can be reconfigured to build a new product within hours.
Conventional robots are impressive machines but they have serious
shortcomings. Because they are often required to assemble several parts, robots
have complex hand-like grippers that can handle a variety of different shapes.
These grippers are attached to equally complex arms, which must be long and
flexible to dip into parts bins and then carry components back to the assembly
line. 鈥淎ssembly-line robots can move in almost any direction because they have
shoulder and elbow joints,鈥� says Hollis.
Advertisement
But the joints that give the robots this flexibility also create problems. No
joint is perfect鈥攁 ball will never fit exactly in its socket, for
example鈥攁nd this limits the accuracy with which a joint will be able to
position the arm. Having more than one joint multiplies the uncertainty. The
inaccuracy of the shoulder joint adds to any deviations in the elbow, wrist or
digits. And as joints wear, the errors inevitably increase. 鈥淭hese are things
that no engineer knows how to compensate for,鈥� says Hollis.
The problem gets worse as the work becomes finer. Hollis has seen smaller and
smaller electronics products come to dominate high-value manufacturing. 鈥淏ut the
robots that put together these assemblies aren鈥檛 shrinking to match.鈥� It is
common to find 100-kilogram robots assembling components weighing only a few
grams. 鈥淚t doesn鈥檛 work,鈥� Hollis says, and consequently up to one in ten
electronic items coming off robotic assembly lines fail to meet quality
standards. Rebuilding poorly assembled goods costs manufacturers millions of
dollars.
So Hollis proposes to do away with programmable robotic arms and their
problematic joints. 鈥淭hen we can achieve precision with far less trouble,鈥� he
says. Instead of a few complicated robots, Hollis鈥檚 factory of the future will
use a large number of simple robots working in concert. These machines will be
one-tenth the size of those now tending automated assembly lines. 鈥淲ith this
approach, we expect to increase precision by a factor of about a hundred. And
reducing the size of robots allows us to reduce the size of a factory based on
them by an equal amount,鈥� he says.
A year ago, Hollis and his 13-person research team at Carnegie Mellon
launched a four-year programme to test the idea. Funded largely by a $2.2
million grant from the National Science Foundation, the team aims to build a
limited prototype assembly line. 鈥淲e鈥檒l build seven or eight robots鈥攅nough
to verify the degree of precision, demonstrate coordination between robots, and
have software for designing and programming the factory,鈥� he says. By the end of
the decade, Hollis hopes to have a demonstration mini-factory capable of
partially assembling one part of a small, complex product such as a hard disc
drive.
Initially, the prototype factory will employ just two kinds of robot. The
first, a courier robot, will fetch and deliver parts and carry the partially
assembled products from one part of the table top to another. The floor beneath
it will be a steel bed composed of tiny square posts spaced 1 millimetre apart.
The couriers will float above the steel table top on a cushion of air and will
steer using on-board magnetic thrusters. The power and air for this process will
be supplied via a cable which also acts as a communications link to a factory
wide internetwork that allows all the robots to communicate.
The second type of robot, which is also linked to the network, is a
stationary manipulator that swivels between a parts bin and the courier robot.
The manipulator simply picks up a component piece, swings over the courier robot
carrying the partially assembled product, and drops down to place the part. To
minimise inaccuracies, it will have only one joint.
The precision with which each part is placed on the subassembly depends on
how accurately the courier can position itself beneath the manipulator. The
courier must therefore know exactly where it is on the surface, and Hollis鈥檚
team is testing two ways in which it might do this. The first relies on a
built-in magnetic detector that senses the metal posts as it glides across the
floor. 鈥淭he robots simply count posts, like counting fence posts going by on the
highway,鈥� says Hollis, and this allows them to monitor their position to within
a millimetre. To get a more accurate fix, the robot measures how the strength of
the magnetic field varies in the gap between each post, and this allows it to
compute its position to within a thousandth of a millimetre.
Fluorescent molecules
The second method uses a fluorescent material embedded in the epoxy resin
between the posts to make a smooth surface. An LED mounted on the courier makes
the molecules fluoresce, and this fluorescent light is picked up through a
series of slits in the bottom of the robot. The robot鈥檚 onboard computer can
then calculate its position as it moves along by reading the pattern of light
created as the slits glide over the posts. 鈥淭he posts block the slits entirely,
partially or not at all,鈥� Hollis says. The courier can use this pattern of light
to determine its position even more accurately.
Either technology will enable a robot to position itself to within a
thousandth of a millimetre of its intended destination, says Hollis, 鈥渁
precision that is completely revolutionary in manufacturing鈥�. Indeed, only human
technicians using tweezers and microscopes come close to achieving such
accuracy.
Each robot will also have other tasks: it must align and orient parts, and
communicate with its colleagues. Neither requires new technology: 鈥淐onceptually,
it鈥檚 fairly straightforward,鈥� says Hollis. 鈥淏ut it will require a good bit of
clever engineering.鈥� Each robot will be designed specifically to hold a
particular part or subassembly, guaranteeing virtually perfect alignment. To
ensure that they grasp and position components accurately, these custom-built
robots will use off-the-shelf robotic vision systems.
While each courier is moving, it will also be chattering away to its
colleagues over the network. For example when two couriers find themselves on a
collision course, they will use the network to negotiate and determine which
should have priority. 鈥淭hese kinds of management questions can be solved with a
central control, but we鈥檙e not doing that,鈥� Hollis says. 鈥淲e want to avoid
having very large programs which become very complicated.鈥� This becomes
especially important as a production line gets more complicated, so that the
centralised program has to become larger still. Hollis is developing an
elaborate set of protocols to allow each robot to make decisions through
consultations with others. He likens the arrangement to city traffic. 鈥淭he
intelligence is contained within each car, not in a master control,鈥� he
observes. 鈥淏ut there are rules of the road to let drivers know how to deal with
each other.鈥�
The network software will give the factory鈥檚 human engineers a voice in the
communications traffic, too. 鈥淭here will be a virtual factory operating as a
three-dimensional computer model showing what鈥檚 going on in the real factory,鈥�
Hollis explains. Operators will be able to monitor each robot or group of
robots, and use the model to reprogram the entire assembly line if they discover
problems or inefficiencies.
Hollis is confident that the modelling software can be crafted from existing
graphics and communication programs. 鈥淏ut there are a number of issues to be
resolved,鈥� he admits. 鈥淒o you execute part of the software at a central site or
all of it locally? What are the best communications protocols to use? All this
needs to be hashed out.鈥�
Virtual collisions
The team is currently tackling these problems using a computer model that
simulates 47 robots assembling a hypothetical product with 34 parts. There is
some way to go. For example, the model does not yet indicate when two couriers
have collided. 鈥淚n our model now, they just drive right through each other,鈥�
says Hollis. He also wants to be able to reroute the couriers by pointing and
clicking with a mouse. 鈥淲e don鈥檛 have that yet either,鈥� he says, though he
expects these controls to be in place by the middle of next year.
The software Hollis is developing is an essential ingredient in his vision
for a future for manufacturing in which assembly plants can spring up, be
reconfigured and disappear in response to shifting market demands and sudden
changes in product designs. 鈥淧roduct life cycles are now measured in months or
even weeks,鈥� he notes. 鈥淭o be competitive, factories have to be able to change
as quickly as product designs do.鈥� Designers will be able to simulate the entire
factory on computer before building it. And manufacturers may not even have to
own the robots they use, but will hire them from specialist suppliers.
Within 10 years, says Hollis, manufacturing engineers will be able to specify
a product design, and then use the Internet to track down the robots that will
be needed to assemble it. 鈥淭hese robots will be sprinkled around the country in
different vendors鈥� inventories,鈥� he forecasts. Using the Internet, engineers
will identify which robots will be needed for the task, download each robot鈥檚
demonstration programs and then simulate the interaction of different robotic
components. This way, the engineers will be able to pick and mix to find the
most efficient combination.
By way of demonstration, he has already opened his group鈥檚 simulated factory
to the Internet, and anyone with a Silicon Graphics workstation can download and
run it.
The minifactory concept could reshape the economics of consumer electronics
industries, says Mark Kryder, a researcher on Hollis鈥檚 team. He cites the
example of computer disc drives, a market that is worth $25 billion a
each year. 鈥淒espite its huge size and rapid growth, the disc-drive industry
suffers from poor earnings because companies can鈥檛 change their manufacturing
processes fast enough to keep up with the changing marketplace,鈥� says Kryder.
The same is increasingly true of other fields. Casio, the consumer electronics
company, calculates that its products typically last less than six months before
going out of production.
For those reasons, manufacturers in Europe and the US are showing a keen
interest in the concept, Hollis says. He hopes that they will eventually fund a
full-scale demonstration factory. But smaller robots don鈥檛 necessarily mean
cheaper ones, he warns. 鈥淭he mechanics are simpler but we double the number of
computers.鈥� However, he鈥檚 convinced that the savings from greater precision will
more than justify the investment.
There is one design factor that Hollis and his team will not change as they
create the table-top robot factory: humans will continue to have a vital part to
play in configuring the factories and maintaining them. 鈥淲hat we propose is
nothing more than a collection of small, simple robots that allows people to
build a custom factory for a specific purpose very rapidly,鈥� he says. 鈥淭his
isn鈥檛 a step toward the personless assembly line. There鈥檚 no replacement for
human intelligence.鈥�