Monday, February 12, 2007

Fly-by-Sight Microrobots


The fly's eye is one of the most well-organized units of visual optics in the world. But is it possible to understand how it works and reproduce it on board neuromimetic robots that could navigate in complete safety? This is the two-part question that Nicolas Franceschini's research team is trying to answer.

The more Nicolas Franceschini tells you about the research he's directing at the helm of the biorobotics department of the Movement and Perception laboratory in Marseille,1 the more you get the urge to stare out the window to watch flies. It's not that there's no buzz in the conversation of this expert in insect sight; quite the contrary. You listen to him enthusing about the amazing behavior of these “agile airships” that he's been studying for over thirty years and that today enable him to design “artificial flying creatures.” These are so efficient, that they make planes and helicopters pale with envy. You just can't help looking out to marvel at the arabesques of these insects buzzing past at high speeds without ever crashing.

But what's so special about the fly's eyes, that have made these insects champion stunt fliers for over 100 million years? First of all, answers Franceschini, “they're totally panoramic, so they can take in the whole of their surroundings.” Futhermore, the fly's “cockpit” contains about 1 million neurons powered by electric signals from the 48,000 photoreceptor cells that make up the mosaic of the retina. The neuron network processes these signals and sends the “electric flight controls” to 18 pairs of motor muscles that adjust the amplitude, frequency, and angle of attack of the wings in real time. This is what lets the fly change direction rapidly, escape from its predators, detect and follow a sexual partner with single-minded determination. The best robots produced today are nowhere close to replicating this kind of precision. Having deciphered how the movement-detecting neurons work, using microelectrodes and special microscopes, the team in Marseille managed to transcribe the main functions into miniature optoelectronic circuits which, 15 years ago, gave sight to non-flying robots that could avoid obstacles without help.

More recently, Franceschini discovered that the insect's retina, which is normally extremely stable, starts to vibrate actively in flight. To understand why this happens, the researchers did a computer simulation of the mechanism, reproduced it technologically and integrated it into a 100-gram twin-engine airship they called Oscar.

“The retina microscanning process allows the fly to keep its eye fixed on a target with a fine-tuning 40 times better than if the retina remained static,” says Franceschini. He's delighted that Oscar never gets flummoxed, in spite of all the devious ways that its designers find to torture it in their experiments (crosswinds, shocks to the fuselage, etc.). Its “vibrating eye,” connected to an electronic movement detector neuron, locks its sight onto the target “like a male fly chasing a female.” This implies possibilities to one day replace both the costly and heavy radars and energy hungry Lidars2 used on some helicopters to locate high voltage cables, by an anti-collision system derived from the fly's eye, that emits no radiation, is inexpensive, and featherweight.

But there's another flying insect-robot with “visuomotor intelligence,” that also has a bright industrial future.

Octave, a 100-gram helicopter, uses an eye in its underbelly that points earthwards, and is directly inspired by the fly's flight-control mechanisms. This allows it to take off and follow a slope at a speed of three meters per second, react to a head wind and land automatically, without any help from avionics (speedometer, altimeter, variometer) but using an “optic flow regulator.”3 This masterpiece could eventually be used as an automatic pilot on spacecraft and aircraft, and even on submarines observing the ocean floor.

These CNRS-patented results, which are at the crossroads of ethology, integrative neuroscience, and robotics do two things at once: They enrich our knowledge of living beings while giving rise to bionic machines inspired by age-old natural principles.

1. CNRS / Université de la Méditerranée joint lab.
The team received the prize “La Recherche 2005.”
2. For Light Detection And Ranging: A detector that emits a laser beam and receives the echo back.
3. “Optic flow” refers to the scrolling of the retinal image created by movement.
The optic flow regulator keeps the optic flow constant by acting on the flying height.

This article is originally done by Philippe Testard-Vaillant for CNRS magazine.
Post a Comment