This project tackles issues of scalability and reality
transfer in evolutionary robotics through the hard
task of scoring goals in robot football.
Evolutionary
Robotics raison d'etre is to allow behavioural
complexity beyond that imposed by the limitations of
design, and as such scalability of techniques to more
complex tasks is a central issue. The use of vision
addresses sensory complexity and a novel technique
of object level representation, using virtual sensors
to provide such a description of the world, is
implemented and seen to work. The use of task
decomposition addresses task complexity, allowing
division of a complete task into simpler
sub-behaviours, making a potentially intractable
problem tractable to solution by the evolutionary
algorithm. Results show that designer
misconceptions can reduce the efficacy of the latter
approach.
Good controllers were consistently
evolved in simulation using genetic programming of
logic-level controllers, and for the most part
successfully transferred to reality. A controller that
could score goals in a real-world environment using
visual location of ball and goal was evolved. The
complexity of this task, within evolutionary robotics
work, goes some way to validating the scalability of
the approaches.
