Robotics Reference
In-Depth Information
around, to manipulate objects) according to a genetically specified con-
troller, while the robot's performance on various tasks is automatically
evaluated. The “fittest” robots are allowed to reproduce by generating
copies of their genetic constitutions but with the addition of changes in-
troduced by some genetic operations (for example, mutations, crossover,
duplication). This process is repeated for many generations until an in-
dividual robot is born that satisfies the performance criterion set by the
experimenter as to its fitness and functionality. Robots might also change
during their lifetime and therefore might adapt to their environment.
The principal components of any evolutionary algorithm are eval-
uation and reproduction, and both of these must be carried out au-
tonomously by and between the robots. The process of evaluation is
carried out autonomously by each robot in the population; for example
a robot that failed to recharge its batteries when necessary would “die”,
or a robot's fitness might be measured in terms of how many objects it
collects within a given amount of time. The selection process in an evo-
lutionary algorithm is carried out by having the fitter robots supplying
“genes”, in which case these robots become parents for robots in the next
generation, or by having the less fit robots lose genes, in which case these
robots are replaced in the population by their offspring. Rather than
pick two parents of above-average fitness and produce an offspring from
a combination of their genes to replace a robot of below-average fitness,
Inman Harvey simplified the genetic algorithm by selecting two individ-
ual robots at random and overwriting some of the genes of the less fit of
the two with some from the more fit of the two.
Robot Offspring, with a Dash of Human Added
Self-reproducing robots will, through evolutionary algorithms, create off-
spring that tend to possess those attributes which the parent robot finds
appealing or even beneficial in itself or in other robots. But this self-
improvement process need not stop there. Your robot might decide that
itlikescertainaspectsof
your
personality, or
your
looks, or
your
voice,
..., and emulate themwhen it reproduces. Your robot might learn a
skill from you and pass this skill on to its offspring. In fact, if it wishes,
your robot could create a part-clone that is more like you than it is like
the robot itself.
