Focus Magazine

S
R
I
N
A W
E R

Y A
0
D
0
F

2 O
SELECTION
Engineering Evolution strategies are
making cars safer
Before computers were readily available, the steel frame of a virtual car to improve its
two pioneers of artiﬁ cial evolution, Ingo fuel efﬁ ciency without reducing safety. The
Rechenberg and Hans-Paul Schwefel, rolled programme varies the metal’s thickness at
dice to incorporate evolutionary strategies over 100 positions and then compares the
into their industrial designs. Through cars to select the ‘ﬁ ttest’ designs. The whole
random mutation and selection, Rechenberg car has to be selected because the various
designed more aerodynamic wings, while thicknesses might only work well in speciﬁ c
Schwefel developed a more powerful nozzle combinations. Successful cars are lighter and
for jet engines. also perform well in a virtual crash simulator.
Now their former student Professor While there is a trade-off between using
Thomas Bäck uses computers to improve his less steel to save weight and using more for
car designs by simulating evolution. Using greater protection, Bäck has still managed to
evolutionary computation, Bäck mutates shave 5kg of steel from BMW’s cars.
An antenna ‘bred’ through
Communications
evolutionary computation
In March 2006, NASA launched three micro- good at sending and receiving signals, while
satellites into orbit as part of its Space being able to ﬁ t in a box just a few
Technology 5 (ST5) project, which aimed to centimetres wide.
test miniature technology for use in future To evolve the antenna, Lohn used a
science missions. During their three months in ‘genetic algorithm’ that uses selection,
operation, the satellites communicated with mutation and recombination – the random
Earth through tiny metal trees. mix and match of genes from two parents – to
The trees are actually antennas, designed produce offspring with new combinations
through evolutionary computation by Dr of characteristics. In the satellite antennas,
Jason Lohn at NASA’s Ames Research Center these characteristics are the various bends
in California. Lohn’s programme starts with and branches of the wire. According to NASA’s
a virtual representation of a crooked piece of lead communications engineer, the evolved
wire and then ‘breeds’ antennas to become antennas worked ﬂ awlessly.
Electronics Evolutionary principles are
As well as improving current designs, the conditions the same. They can be used
leading to innovative electronics
evolutionary computation is also used for a wide range of applications, from the
to innovate. Dr John Koza, a consulting thermostat in your home to the cruise control
professor to Stanford University, has used in your car.
evolutionary programming to re-invent over Processing power is the major limiting
a dozen electrical circuits and converters. His factor to evolving inventions. A decade ago it
programmes have created designs that are would take weeks to duplicate 20th-century
similar to – yet don’t infringe upon – existing electrical circuits, but now our computers
patented designs. are powerful enough to re-invent more
One invention, a type of controller, has complicated 21st-century electronics, and
even been granted its own patent. Controllers in less time. And because processing speed
monitor and affect the conditions of a doubles every 10 years, it’s likely we’ll see
system through a feedback loop to keep more evolved inventions soon.
Robotics
Cornell University’s Professor Hod Lipson furry blocks that evolve in response to random
builds machines that learn to walk. But while mutations. Unlike most applications of
most machines learn by trial-and-error, evolutionary computation, their programming
Lipson’s robots do it by ‘thinking’. isn’t rewarded for achieving an objective. As
The Starﬁ sh robot begins by moving each the blocks evolve, one in 10 collect others,
arm a little, to help it understand how actions forming a multi-block organism that moves
affect its movements. This lets the robot build around before splitting into offspring.
up a mental image of itself that it can use to The robots are therefore self-replicating,
plan a way of moving, without having to test suggesting that when things evolve in the
whether it will work. This seems to be how absence of a reward, their objective becomes
The Star sh robot
intelligent animals decide whether a method replication. The replicating blocks might shed
learns to walk
will succeed without needing to try it ﬁ rst. light on a question that bafﬂ ed even Darwin:
Lipson has also created a robotic population of what is the origin of life?
www.bbcfocusmagazine.com February 2009 41
Focus 199 darwin1_10RB.indd 41 18/12/08 15:24:15
Page 1 | Page 2 | Page 3 | Page 4 | Page 5 | Page 6 | Page 7 | Page 8 | Page 9 | Page 10 | Page 11 | Page 12