Self Constructing Neural Network Robot Controller based on On-line Task Performance Feedback

A novel methodology to create a powerful controller for robots that minimises the design effort is presented. We show that using the feedback from the robot itself, the system can learn from experience. A method is presented where the interpretation of the sensory feedback is integrated in the creation of the controller, which is achieved by growing a spiking neural network system. The feedback is extracted from a performance measuring function provided at the task definition stage, which takes into consideration the robot actions without the need for external or manual analysis.