91av

YELLOW and black blobs streak past me with centimetres to spare, careering
about with an effortless precision that would shame even the most daring of
stunt pilots. One makes a grazing landing on my head, before buzzing away, doing
a back flip and landing on the ceiling. Another singles out my nose, hovers
within a whisker of my left nostril, then flits off and zooms neatly down the
dead centre of a long perspex tunnel, for all the world like an X-wing fighter
zooming through the conduits of the Star Wars Death Star.

Such airborne agility is especially remarkable when you consider that the
honeybee does it all on a shoestring, neurologically speaking. Its brain is only
as big as a sesame seed, which is less than a millionth the size of the human
brain. Its eyes—arrays of hundreds of individual lenses, each looking in a
different direction—are also far simpler than ours.

That honeybees and other insects accomplish so much with so little hasn’t
been lost on the likes of NASA and the US and Australian departments of defence.
That’s why I’m visiting the All-Weather Beeflight Facility at the Australian
National University here in Canberra. This glorified greenhouse is home to a
vital part of an international effort to create the next generation of Mars
probes and military spy robots. The plan is to build tiny flying craft that can
navigate autonomously, seeing, thinking and avoiding danger using the same
tricks insects have for over 350 million years.

As I adjust to the balmy conditions inside the bee house, behavioural
scientist Shaowu Zhang explains how you persuade a bee to fly down a perspex
tunnel—fairly easily, it turns out. Zhang first lures the bees to the
tube’s opening by placing a drop of sugar water there. As the bees zoom back and
forth between the hive and the sugar, he moves the sweet reward further and
further into the tunnel until the bees are flying all the way to the the far
end. The whole process can take just three hours. Once the bees are trained,
there’s ample opportunity to experiment on them—worker bees may make as
many as 90 visits a day to a source of sugar, covering up to 30 kilometres.

But it’s painstaking for the humans involved. Two graduate students sit
beside one of the tunnels, oblivious to the maelstrom of stinging insects. As a
bee enters the tunnel, one of them shouts out: “Bee number 18,” while the other
notes it on a clipboard. Sure enough, I see two minuscule dabs of
paint—blue on the thorax and orange on the abdomen—representing the
number 18.

For a moment I question the sanity of these people. Don’t we already have
perfectly good navigation systems, without chasing bees down tunnels and dabbing
their rear ends with paint? Well, yes and no.

There’s the Global Positioning System, or GPS, which guides some versions of
the Tomahawk missile, but relies on signals from satellites, which an enemy can
jam. What’s more, you can only use GPS to navigate around features that have
already been mapped. If you’re a miniature spy craft flitting through an urban
jungle, GPS won’t stop you slamming into a rubbish bin or getting tangled in a
clothes line. And of course, GPS is available only on Earth, so it’s no good for
exploring Mars.

Other navigation systems detect obstacles by bouncing sound, lasers or radar
off them. Trouble is that the signals they emit can betray you to the enemy and
get you blasted into confetti. All in all, there’s plenty of room for
improvement.

So what does it take to flit about as nimbly as a bee? The secret to steering
a safe course lies in being able to judge distances. If you’re a bee, you need
to know whether that’s a rubbish bin on the other side of the room, or a salt
cellar you’re about to slam into. Bees do this differently from humans.

Like most vertebrates, we have binocular vision—each eye sees an object
from a slightly different angle, resulting in two images with subtle differences
that allow the brain to estimate distance. How well you can do this with
binocular vision depends on how far apart your eyes are. This is why praying
mantises, the only insects known to use binocular vision, can only judge
distances up to a couple of centimetres away.

What’s more, binocular vision requires a lot of brain power. That makes it a
poor starting point for designing energy-efficient craft to explore faraway
planets, or tiny undercover spy planes to snoop around your enemy’s
headquarters. Luckily, insects like the honeybee have a far simpler strategy for
seeing 3D—something called optic flow.

Mandyam Srinivasan (aka “Srini”), an electrical engineer turned
neurobiologist had explained optic flow to me over the phone before my visit to
the bee facility. In a nutshell, the closer things are, the faster they appear
to flow by as you move.

As I sat on the train speeding towards Canberra, hoping to catch a glimpse of
a wombat or wallaroo outside, it was easy to see what he was talking about. The
silhouettes of gum trees far off on the horizon crept by slowly, while
farmhouses in the mid-ground marched along at a more industrious pace. And the
sheep fences close to the track positively raced past, 10 or 15 fence posts a
second. But while we humans use optic flow merely to supplement our binocular
vision, honeybees depend on it.

As Srini leads me through the bee house, he tells me how an insect might use
optic flow to manoeuvre through a cluttered environment such as dense
vegetation. “If the insect suddenly finds something is moving very rapidly on
its left side,” he says, “it veers away from it, because it knows there’s an
obstacle there.”

We pause in front of a tunnel whose interior walls are covered with a
polka-dot pattern. Through the perspex roof we see a steady stream of bees
buzzing precisely down its centre. “What they’re doing,” he says, “is simply
balancing the velocity of the image on the two sides.”

To illustrate his point, he flips a switch. A belt turns, and the polka-dot
pattern on one side of the tunnel begins to move in the direction that the bees
are flying—so relative to the bees the pattern has slowed down. Slower
means further away, so the bees veer toward the moving wall, hugging it tightly
as they whizz down the tunnel.

Srini and Zhang have done numerous experiments using the tunnel set-up over
the past decade, and they all point to the same conclusion. Honeybees are
programmed to keep the speed at which images flick over their eyes constant
during flight. For instance, when both walls move in the same direction as the
bee, it flies faster. “It automatically ensures that you speed up when you’re in
a wide open space and slow down when you’re in a cluttered, more dangerous
environment,” says Srini.

If you’re a bee, optic flow is also essential for landing
(91av, 29 July 2000, p 13).
Anyone who’s ever fallen off a moving bike
or horse knows that sickening feeling as the closer you get to the ground, the
faster it seems to be hurtling by. Unlike us, honeybees slow down automatically,
altering their flying speed to keep the speed at which the ground rushes by
constant, while at the same time descending at a constant angle. “It’s like
being on autopilot,” says Srini, “you don’t need to know your speed, you don’t
have to know your altitude. It does the job for you.”

Honeybees also use optic flow to tell them how far they’ve travelled. When
Zhang trained bees to fly halfway down a tunnel to find some sugar water, he
found he could confuse them by switching the tunnel for a wider one: on return
flights the bees overshot. The bees translate this distance information into the
waggle dances they use to tell their hive-mates where the best nectar sources
are (Nature, vol 411, p 581).

So what does all this mean for Mars probes and flying robotic spies? Well,
within a decade optic flow could help the craft steer a safe course. At first,
they’ll still be partly dependent on remote control. Optic flow will take over
only for the simple stuff like avoiding unexpected obstacles. Eventually, truly
autonomous robots will emerge. They will use optic flow to negotiate winding
canyons or corridors as deftly as bees, avoid obstacles in cluttered
environments, and even land to pick up samples, predicts Steven Zornetzer,
director of Information Sciences and Technology at NASA’s Ames Research Center
in Moffett Field, California.

Spaceflight and spying

These flying robots would be packed with a battery of miniaturised
instruments. A locust-sized robot fitted with a camera could check out the
inside of a suspected biological weapons plant, or sniff out rebel ambushes
along a strategically important road. Or a Mars probe could could go off looking
for water—and hence potential signs of life.

To fully comprehend the advantages of an autonomous, self-steering robot,
think back to 1997 and the Mars Pathfinder mission. It was a huge success, even
though the rover exploring the planet’s surface moved at a snail’s pace,
travelling just 52 metres in 30 days. This was because it got its instructions
from Earth, 190 million kilometres away—so each command took 11 minutes to
travel between the two planets. Move the rover any faster and it might have
ended up upside down at the bottom of a gully with its wheels spinning in the
air before mission control even knew it was entering rough terrain.

Imagine instead that Mars Pathfinder had released a swarm of autonomous
flying robots to scope things out. “That would be a real exploration mission,”
says Zornetzer. “They could look for erosion patterns indicating past presence
of water, or striations in vertical cliffs, or sedimentary deposition, and send
all that information back to Earth.”

Srini also has ideas of how optic flow could change lives at a more
day-to-day level. Optic-flow detectors connected to a beeper could be fitted to
canes for blind people, he muses. “By moving the cane from side to side, you
could measure the optic flow generated by obstacles in front of you.” Traffic
police could nab speeding cars with optic-flow sensors hidden in road signs,
without drivers being able to spot them with a radar detector.

Before these dreams can be realised, however, researchers have still to
perfect the computer algorithms that will allow the robots to navigate via optic
flow. Srini is tackling this problem together with Javaan Chahl, a roboticist
from the Australian Defence Science and Technology Organisation. The two
researchers test successive generations of those algorithms using an “eye”
mounted first on a robotic arm that manoeuvres through several square metres of
papier mâché mountains, and then on a specially designed
1.5-metre-long helicopter.

Remember that typical insect eyes wrap around the head, providing as good a
view of what’s behind the insect as what is in front. Not to be outdone, Chahl
and Srini’s robotic eye comprises a camera that looks down onto the vertex of a
conical mirror, capturing a complete, albeit distorted, 360° view of the
camera’s surroundings.

With an ordinary camera, says Chahl, “you might see everything moving to the
side, but you won’t know whether you’re moving sideways or spinning”. That’s not
the case with the robotic eye. When I spin on my stool, looking down on the
conical mirror, I can see the whole world revolving about me uniformly. If I
scoot side to side, one edge of the image contracts while the other expands.

Using this robotic eye linked up to the optic-flow algorithms, the helicopter
can hover on the spot, drifting only 5 centimetres a minute. That may sound
unsexy, but it’s actually a major achievement. Indeed, Chahl’s best
remote-controlled efforts look decidedly inebriated in contrast.

Hovering may be more difficult, but it’s forward flight guided only by optic
flow that thrills the naive observer like myself. This exercise is conducted in
a field safely removed from civilisation—and no wonder. Watching the test
flight, I’m reminded of a farmer chasing an escaped bull. The helicopter skims
smoothly along at 70 kilometres per hour, while a pickup truck rattles after it
in hot pursuit. A researcher perches in the back, ready to take over with remote
control if things go wrong. They don’t.

That’s a far cry from being able to fly through winding canyons, or low over
a series of mountain ridges, but already, the helicopter’s achievements have
attracted the attention of military engineers. “For the helicopter to control
itself and hover, that’s extremely difficult. It really hasn’t been done before
except under extremely controlled conditions,” says Johnny Evers, a systems
engineer at the US Air Force Research Laboratories Munitions Directorate at the
Elgin Air Force Base in Florida.

While Srini’s team works to perfect its optic-flow algorithms, other groups
tackle the next challenge: miniaturising all the technology and making it
reliable enough to build what Chahl calls “the dream”: a robotic aircraft with a
one-inch wingspan.

For a start, the robots will need back-up sensors to help them steer when
optic flow is not enough. So they’ll also have gyroscopes, airspeed and inertial
sensors, horizon detectors, compasses, and so on. Flying insects also have their
own versions of these sensors
(see “How bees steer a safe course,”), so
no prizes for guessing where scientists are turning for inspiration about how to
integrate all those systems.

Michael Dickinson and Ronald Fearing at the University of California at
Berkeley are in the early stages of building an insect-sized Robofly. Just like
the real thing, it will use an “intelligent shell”—the input from each of
the insect’s sensors directly modulates the sensitivity of others, and it
manoeuvres largely by involuntary reflexes, almost no matter what.

“Insects can stay in the air with torn wings, missing legs, blind eyes or
when burdened with twice their weight in additional mass,” says Dickinson. A
Mars probe fitted with similar technology would be able to land safely even if
mission control screws up and confuses metric and imperial units.

Defeated by decor

Good back-up systems will also be vital in bad weather and for certain indoor
environments. Optic flow depends on tracking the motion of edges—the
boundary between light and dark areas. “You need contrast,” says Nicolas
Franceschini, a bioroboticist at CNRS, France’s national research agency in
Marseilles. “Fog and rain blur the contrast terribly,” he says.

A moonless night, however, is unlikely to be a problem. There is usually
enough contrast, once modern image-intensification software has got to work on
the input. Indoor environments are also fine—as long as the enemy goes in
for shag carpets, stucco and flock wallpaper. “If you have low contrast, like
glass windows or even nicely painted walls, you’re lost,” says Franceschini.
This is why bees sometimes end up hammering away at walls and windows, unable to
orient themselves.

Back-up systems will be a moot point, however, unless Srini and his team can
also scale down the size and energy consumption of the robotic eye and its
optic-flow computer to honeybee proportions.

At the moment, the optic-flow helicopter uses power-hungry Pentium chips. A
better option, according to engineer Charles Higgins at the University of
Arizona in Tucson is his specially designed chip that better mimics the bees’
energy-efficient design. The chip, which Srini’s team will soon try out,
incorporates both the robotic eye and the computer processing, and consumes
almost no energy while idle. This makes for a 100-fold reduction in power
consumption, and a 10-fold reduction in weight.

Then there’s the monumental task of making the wings and motor
honeybee-sized. Gears and pulleys don’t work well at such tiny scales, so
Dickinson is trying to mimic insects’ aeronautical mechanics—Robofly will
have no rotating parts, and will flap its wings by vibrating its
exoskeleton.

Of course, the ancestors of humans and insects went their separate ways more
than half a billion years ago, and so wide an evolutionary gap is not bridged in
a few short years. “Our helicopter weighs about seven or eight kilos. That’s a
long way to seven or eight grams,” admits Chahl.

But as I stand in the bee house, about to brush off a bee that’s exploring my
head, and pondering just how far Srini and his colleagues have already come,
it’s hard to believe they won’t succeed as long as they stick to nature’s
blueprint.