| Popis: |
In controller synthesis for basic robotic assembly tasks, the optimal performance is characterized by high level criteria. Performance of a peg-into-hole task is, for example, measured in insertion time and average/maximum force level. Moreover, the unknown optimal control of a peg-into-hole task was shown to be nonlinear. Fuzzy rules are used in our approach to approximate this nonlinear control. The fuzzy controller synthesis can be automated with the use of a machine learning tool SMART+, provided that examples are available. Besides examples, SMART+ can also handle domain knowledge as input. Its output can also be a learning fuzzy controller, in order to achieve online performance improvement. The paper presents this approach and initial experiments. > |
