Toward Evolved Vision-based Control for a Quadrocopter
Szczawinski, P. S.
; Duarte, M.
Toward Evolved Vision-based Control for a Quadrocopter, Proc Conf. on Telecommunications - ConfTele, Castelo Branco, Portugal, Vol. --, pp. -- - --, May, 2013.
Digital Object Identifier: 0
In this paper, we show how evolutionary robotics
can be combined with computer vision in order to create a
controller for a ﬂying robot. In our experiments, the robot is
required to stay over a speciﬁed area using visual data from its
downward pointing camera. The robot’s controller is an artiﬁcial
neural network, whose weights are optimized by an evolutionary
algorithm. The task and the robot are modelled in simulation,
where controllers are evolved. After the evolutionary process is
concluded, the best controller is transferred to a real robot. We
validate our approach by showing that the evolved controllers
are able to execute the task not only in simulation, but also on
the real robot.