[tt] NS: Shape-shifting robots take form Movie Camera

Mon Apr 28 18:05:08 UTC 2008

Shape-shifting robots take form Movie Camera
http://technology.newscientist.com/article.ns?id=mg19826531.200&print=true
25 April 2008
Jeff Hecht

How would you like to have your very own shape-shifter? Perhaps a
liquid metal T-1000 Terminator to help around the house. Or a
universal tool kit that could reshape itself into any implement at
the press of a button. For an astronaut in orbit, an army mechanic
in remote terrain or even a homeowner trying to fix a furnace on a
cold winter night, it could be just the thing.

Well, one day maybe. The traditional approach to building
shape-shifting devices has been to use materials based on shape
memory alloys, polymer sheets or nanoparticles. But these have
proved difficult to control and have other limitations, so
researchers have begun taking a different and less exotic tack.

Their approach is known as self-reconfigurable robotics, and it
takes advantage of recent advances in robot hardware, communications
and control algorithms. Last year a team of US researchers from
different universities put together a blueprint for shape-shifting
based on cell-like robotic "modules" that can rearrange themselves
to create different shapes. They made their pitch to the Defense
Advanced Research Projects Agency. As a result, DARPA set aside $4
million for a set of six-month studies to design modules that could
be mass-produced for demonstrations. The agency is prepared to fund
18-month contracts to pursue the most promising designs.

Instead of trying to control individual molecules or create
nanoparticle fluids that morph from puddles to silver-skinned
cyborgs, DARPA programme manager Mitch Zakin is pursuing what he
calls "programmable matter". These are so-called "mesoscale"
mini-machines, a millimetre to a centimetre in size, that can
arrange themselves to form whatever shape is desired. Initially,
Zakin expects the outcome to be devices the size of small Lego
pieces, but as the technology improves the modules and the machines
assembled from them should scale down further. Ultimately you could
tell a sack of "smart sand" what to do, and the grains would
assemble themselves into a hammer, a wrench or even a morphing
robotic aircraft. "It's making machines more like materials, and
materials more like machines," says Daniela Rus, a robotics
researcher at the Massachusetts Institute of Technology.

The current era of reconfigurable robots began in 1988, when Toshio
Fukuda of Nagoya University in Japan designed a "cellular robot"
that could change its form. In the 1990s, roboticist Mark Yim, then
at the Palo Alto Research Center in California, built a snake-like
robot that could be put together in different configurations to move
around in different ways. Each part contained a microprocessor,
sensors and a motor to allow it to move relative to its neighbours.
Then in 2002, computer scientists Seth Goldstein and Todd Mowry of
Carnegie Mellon University in Pittsburgh, Pennsylvania, began using
on-board electromagnets to move modules around and get them to stick
together in an approach they called "claytronics" (New Scientist, 11
June 2005, p 30).

Generations of graduate students and researchers have cobbled
together similar demonstrations, but few have progressed beyond
slow-moving modules the size of toy blocks. "We haven't made much
progress in two decades," admits roboticist Hod Lipson of Cornell
University in Ithaca, New York. For his part, Lipson has
demonstrated robots that assemble themselves from simple components,
and he thinks this is where the future lies. He envisions millions
of mobile modules that can join together like living cells to make
finely granulated structures.

Last year Zakin decided, after talking with researchers including
Lipson, that DARPA should try to help move things in that direction.
At a systems and technology symposium last August, Zakin spoke about
modules that would couple, uncouple and rearrange themselves into
new shapes and structures. They would transfer information and
energy among themselves, and have the internal smarts to respond
almost organically to their environment. To do this, each module
would require its own motor to move it, an electromagnet or other
means to hold on to other modules, and a microprocessor to
coordinate its motion (see Diagram).

The first challenge is to provide the modules with an efficient way
to align themselves relative to each other. Yim, who is now at the
University of Pennsylvania in Philadelphia, has developed a
promising method. He is working with modules made up of a pair of
metal pieces linked by a motorised hinge. Plates on the side and
bottom of these modules have magnets that can be used to join them
to adjacent modules in any of four orientations. Some, but not all,
modules also contain a digital camera and processor. The advantage
of using cameras as sensors is that they use very little power, he
says, and they don't interfere with each other.

Yim gives a dramatic demonstration, which he calls the first case of
"self-assembly after explosion". It starts with a 15-module robot
walking on two legs. A student kicks it apart and it falls into
three pieces, each made up of four "ordinary" modules and one camera
module. The pieces wiggle across the floor towards one another,
guided by images from the camera and computer-vision algorithms.
They then reconnect, re-forming the original robot, which slowly
rises and walks on.

A device that can put itself back together like this is all well and
good, but an important question remains: how should shape-shifting
modules be arranged in the first place? The simplest design is a
snake or chain - which has been demonstrated in many labs - with
each module attached to one or two others. By twisting the segments
back and forth, an assembled snake can slither across the floor.
Some variants also have arms and legs: Yim recently assembled a
record 56 modules into a 14-legged robot that can walk. He claims
that the control system would allow even more modules to be added,
but in general a snake design doesn't scale up well to thousands of
modules. That's partly because it would be hard for an on-board
computer to keep track of the many directions of movement, but the
main difficulty is that the control signals have to pass from module
to module, along the length of the entire structure.

Crystal morph

To get around this problem, researchers are experimenting with an
architecture in which modules are arranged in a lattice-like 3D
pattern, a bit like atoms in a crystal. The lattice offers parallel
paths for control signals, so different elements in the lattice can
move simultaneously. The regularity of a lattice also makes it
easier to rearrange or model on a computer than a chain, says Mark
Moll, a computer scientist at Rice University in Houston, Texas. All
in all, he says, the lattice design should scale up to large numbers
of modules more easily.

Take the "Miche" system that Rus has developed at MIT. It is a
self-assembling robot with a twist: instead of building a robot
structure by adding modules, Miche starts with a large lattice and
removes modules, like a sculptor chipping into a block of marble.
Rus begins by stacking together 26 cubic modules, each measuring
less than 5 centimetres to a side. When stacked, the robotic modules
communicate with their neighbours through infrared sensors on their
faces. Switchable magnets hold the lattice together.

Rus assembles the lattice on a table and lifts it into the air.
While the electromagnets are switched on, the pieces hold together,
but when control signals turn off the magnets holding a particular
block it drops to the table. As more blocks drop away, a dog-shaped
robot emerges. The experiment makes the point that to get a shape
you want, it may be simpler to remove modules from a lattice
structure than to add them or move them around.

Although researchers have not yet assembled more than a few dozen
modules, they are making plans to build more complex structures.
Assembling one module at a time into a predetermined spot,
autonomously or by hand, simply won't work for millions of modules.
It would take too long and modules would wind up in the wrong
places. "You have to give up the idea of deterministic
manipulation," Lipson says. Instead he suggests a probabilistic
approach, in which large numbers of modules are shaken up so they
end up filling a desired volume. Individual modules wouldn't have to
find their way to assigned slots; instead, they could adapt to the
niche they find themselves in, based on information from other
modules and an overall control system.

Electrical engineer Eric Klavins of the University of Washington in
Seattle has demonstrated this idea by shuffling "programmable parts"
on an air-hockey table. Each module is an equilateral triangle 12
centimetres to a side, with a magnetic latch on each face. When two
triangles collide, they latch together while their on-board
processors communicate with each other via an infrared link to
decide whether or not to remain attached. In this way Klavins can
pre-program the triangles to build up complex structures such as
parallelepipeds and hexagons. His physical demonstrations have been
limited to a small number of modules, but his computer models have
extended the idea to larger scales.

Ultimately his goal is not to specify the behaviours precisely, but
rather to make the structures assemble themselves. Klavins is
inspired by the way molecules form living cells and living cells
form tissues, and he aims to glean lessons from nature to apply to
robotic modules. He is now trying to build complex structures using
simple, cheap materials by designing them from the bottom up. "The
fundamental ideas are very similar for programming proteins and
programming structures," he says. "But I don't know what kind of
technology we'll be able to use."

No matter how it is achieved, any big change in scale will raise
problems that so far have been addressed only in theory. Control and
communications with millions of modules is a huge challenge, and
special algorithms will be needed to decide how much information
individual modules need. "How do the modules talk to each other to
come to a consensus and agree on a series of actions?" Moll asks.
Will all modules be identical, or will different modules be needed
for different tasks? How will energy be distributed among millions
of modules?

Then of course there is the enormous engineering challenge of
shrinking the size of each module. Even stepping down to Zakin's
mesoscale poses serious hurdles. Shrinking motors is very difficult,
because they have to be both strong and energy-efficient. One
approach, Yim suggests, is to use external energy - shake the system
to get modules to settle into place - but that would shift the
problem to control and sensing, which will still need to be worked
out on the mesoscale.

So what lies ahead for programmable matter? DARPA issued its request
for proposals last October, and in the next two years developers aim
to demonstrate a system with at least 1000 modules - an ambitious
20-fold increase in complexity over the biggest system to date. A
universal tool kit is further off, but at least it won't have to
compete with everyday hardware in terms of costs. The Pentagon would
gladly pay a premium for a single item that saves soldiers from
hauling a truckload of special-purpose tools into combat, just as
NASA would pay for one that saves weight on space missions.

For practical uses, the overall strength and robustness of any
robotic tool will be crucial. DARPA asks only that the first round
of implements have the strength of plastic, but that won't do for a
universal tool kit. And how will robotic modules deal with the
inevitable problems faced by field equipment such as dirt and
misalignment? Questions like these will have to be answered.

If the project succeeds, a universal tool kit will be just the
start. Robotic modules that can rearrange themselves could also
repair themselves by moving modules to replace damaged sections.
Imagine a vehicle that could repair its own dents and scratches, or
clothing and other gear that could recover from wear and tear.
Further down the road, reconfigurable robots might learn to make new
modules from raw materials, thereby becoming self-replicating.
Perhaps shape-shifting robots will become the new explorers,
plumbing the depths of the ocean and exploring planets too hostile
for humans.

We have a long way to go, but even a child can grasp the universal
appeal of shape-shifting. Rus says that when her daughter was 4
years old, she said she wanted to be able to "pull toys out of the
wall". The girl wanted her own universal toy kit. If she doesn't
grow up too fast, maybe she'll get one.

Robots - Learn more about the robotics revolution in our continually
updated special report.

Related Articles

Shape-shifting robot forms from magnetic swarm
http://technology.newscientist.com/article/dn13244
29 January 2008
Shape-shifting 'smart dust' may explore alien worlds
http://technology.newscientist.com/article/dn11624
18 April 2007
Armed and ready
http://technology.newscientist.com/article/mg16922793.200
24 February 2001
Atomic logic: In search of shape-shifting circuits
http://technology.newscientist.com/article/mg19726422.000
09 February 2008

Weblinks

Mark Yim, University of Pennsylvania
http://www.grasp.upenn.edu/people/yim.html
Claytronics project, Carnegie Mellon University
http://www.cs.cmu.edu/~claytronics/
Daniela Rus, Massachusetts Institute of Technology
http://groups.csail.mit.edu/drl/wiki/index.php/Main_Page
Eric Klavins, University of Washington
http://soslab.ee.washington.edu/mw/index.php?title=Main_Page

Users Demand Expertise at How-To Web Sites
http://www.nytimes.com/2008/04/28/technology/28ecom.html

E-mail me if you have problems getting the referenced articles.