Researchers at Sandia National Laboratories led by computational neuroscientist Frances Chance are looking to the common dragonfly for clues to develop smaller, more efficient missile defenses. By replicating the predatory insect's brain in a computer algorithm, the goal is to create interceptors that can lock on to threats much faster and achieve a much higher kill ratio.
Dragonflies have been around for about 325 million years and haven't changed all that much since then, so they must be doing something right. Part of the reason they've lasted so long is that, though they are typically associated with bucolic moods as they buzz about on warm summer evenings, they are one of nature's consummate predators with a kill ratio of 95 percent once they've targeted their prey.
The dragonfly manages this through its remarkable brain, which on first inspection seems to be a very simple, even primitive thing, but is capable of some remarkably fast and complicated computations. When it pursues its prey in flight, the dragonfly doesn't chase after it. Instead, it anticipates where its dinner will be and calculates a straight intercept course that it corrects as its target bobs and weaves.
This is pretty good, considering that the dragonfly doesn't even have depth perception, so how does it do this? To find out Sandia National Laboratories did some reverse engineering based on the real dragonfly's behavior and created simulated dragonflies in a digital environment that duplicated the insect's brains as neural networks.
According to Sandia, the results mimicked the dragonfly brain with great accuracy. This is of great interest because the dragonfly can react to its prey in only 50 milliseconds, or six times faster than the blink of a human eye. Since this is the time that it takes a signal to pass through a mere three neurons, each dragonfly computation must take only three steps, though living brains do carry out a kind of parallel processing, so a lot of calculations can be done in a very short time by a very simple set of neural circuits.
By contrast, conventional missile defense systems use much more computing power for a very similar task. By using the dragonfly brain as a model, it may be possible to make smaller, lighter computers that need less power to operate, as well as increasing kill ratios. In addition, the dragonfly algorithm could help to intercept less predictable hypersonic missiles or show how to calculate interceptions using less sophisticated sensors.
The researchers concede that there are fundamental differences between dragonflies and missiles – speed being the most obvious. However, even if missile defense proves to be a bust, the new technology could be of great use in artificial intelligence and applications like self-driving cars or prescription drug development and testing.
The research will be presented by Chance at the International Conference on Neuromorphic Systems running this week in Knoxville, Tennessee.
Source: Sandia National Laboratories
