Abstract
We introduce a novel approach to evolution of robot
control programs in simulation, generalised to a real,
physical robot, where the remaining step in evolution
is carried out. First, we briefly describe an
evolutionary programming experiment performed with our
biped humanoid robot elvina, with onboard computer and
sensors. By using this system, we evolved gait patterns
that outperformed the previously manually developed
gait. Second, the physics simulator used is presented.
It is a free, industrial standard library for
simulating articulated rigid body dynamics, designed
for use in interactive or real-time simulation of
moving objects in changeable virtual reality
environments. The simulation is based on a method where
the equations of motion are derived from a Lagrange
multiplier velocity based model and it uses a highly
stable, first order integrator. Finally, we present the
Genetic Programming system used for evolution.
Evolutionary algorithms mimic aspects of evolution and
Darwins principle of natural selection and survival of
the fittest, in order to optimise a solution towards a
defined goal. The primitives of Genetic Programs are
the terminals and the functions. The terminals are
comprised of inputs to the program, constants or
functions without arguments. The functions are composed
of the statements, operators and functions available to
the GP system.
Users
Please
log in to take part in the discussion (add own reviews or comments).