On-Line Learning Using a Connectionist Controller for Simulated Robot Soccer

Robot soccer is a dynamic, real time robotic task that is commonly used as a study platform in many areas of computer science and computer engineering. This paper presents a robot soccer controller based on an attentional mode neural network. This controller learns incrementally using a single training episode and does not require any kind of previous off-line learning. The performed experiments show that the controller presents a good generalization, accomplishing the task even in situations not previously experienced.