Programming an Autonomous Robot Controller by Demonstration Using Artificial Neural Networks

The use of artificial neural networks (ANNs) to control autonomous robots has been quite extensively studied. Also, in recent years researchers have begun to investigate the notion of programming such robots using visual programming control models. Some of this work has focused on developing languages based on various programming and robot visual programming-by-demonstration (PBD) systems. Here we extend the latter approach by proposing a visual PBD environment for autonomous robots based on ANNs. Within this environment, sensor-to-motor rules, called sensorimotor maps, are programmed by employing ANNs to match sensor outputs to actuator inputs. The goal is to create a programming environment in which the end-user is not required to have any knowledge of the underlying control model, ANN programming in this case. In this regard, the current proposal appears more promising than previous attempts using the subsumption model