MAKING locusts watch battle scenes from Star Wars could eventually
help cars steer themselves out of trouble.
Researchers in Britain and Switzerland have already exploited this bizarre
research to build a robot that can dodge approaching objects. They hope the
robot鈥檚 neural network program, which mimics part of the locust鈥檚 brain, will be
the basis of a collision-avoidance system for cars.
The conventional approach to creating such systems involves using radar or
infrared detectors, and requires very heavy-duty computer processing. Insects,
however, manage to dodge objects despite their poor vision and basic brains.
Advertisement
In locusts, the key to this ability is a large neuron behind each of their
compound eyes called the lobula giant movement detector. The LGMD is thought to
be partly responsible for triggering the insects鈥� escape jumps and steering
responses during flight, says Claire Rind, a neurobiologist at the University of
Newcastle.
So copying the behaviour of the LGMD could help create fast and efficient
avoidance systems for vehicles such as cars and planes, Rind says. 鈥淭he
advantage of using this neuron is that it can discriminate between objects that
are on a collision course and ones that aren鈥檛,鈥� she says.
To build up a picture of how this neuron works, Rind showed a locust images
of rapidly approaching objects while monitoring the neuron鈥檚 activity. She chose
scenes from Star Wars, she says, because the film has particularly good
shots of objects鈥攕uch as Tie fighters鈥攈eading for the viewer during
dogfight sequences.
Working with Mark Blanchard and Paul Verschure of the Institute of
Neuroinformatics at the Swiss Federal Institute of Technology in Zurich, Rind
then built a robot with a vision system based on the locust鈥檚. The small
three-wheeled robot has a camera with a resolution of just 20 by 20 pixels,
which closely imitates the locust鈥檚 grainy vision.
As in the locust, signals from these makeshift photoreceptors are subjected
to some pre-processing by a neural network鈥攑roviding some basic 鈥渆dge
detection鈥濃�攂efore they reach the robot鈥檚 movement detector, which is
designed to respond in the same way as the LGMD. The researchers found that
their robot avoided objects 91 per cent of the time, even though it had only a
short time to react.
Nigel Clarke, principal engineer at the car company Jaguar鈥檚 radar
applications research unit in Coventry, has been keeping a close eye on Rind鈥檚
progress. 鈥淲e certainly think there鈥檚 a great deal of mileage in using
techniques that the brain uses to calculate these things, such as detecting
moving objects in a scene,鈥� he says.
Blanchard believes that traditional approaches to collision
avoidance鈥攕uch as infrared or laser rangefinders and ultrasonic
reflections鈥攁re too easily confused by spurious signals. 鈥淢ost biological
systems can see,鈥� he says, 鈥渟o it鈥檚 at least one approach that should be
exploited in trying to make machines explore for themselves.鈥�
But Rind says more research is needed before the technology can be exploited
commercially. She now wants to monitor the locusts鈥� response to natural stimuli,
instead of movies, and work out how the speed of approaching objects affects
behaviour.