The authors present a neural network approach to solve the dynamicscheduling problem for pick-place operations of a robot-vision-trackingsystem. An optimal scheduling problem is formulated to minimize robotprocessing time without constraint violations. This is a real-timeoptimization problem which must be repeated for each group of objects. Ascheme which uses neural networks to learn the mapping from objectpattern space to optimal order space offline and to recall online whathas been learned is presented. The idea was implemented in a real systemto solve a problem in large commercial dishwashing operations.Experimental results have been shown that with four different objects,time savings of up to 21% are possible over first-come, first-servedschemes currently used in industry
展开▼