Skill acquisition from experimental data

The authors present a new bottom-up approach to acquire skills in control. The acquired skills represented in the form of state transition laws are modeled by using a Petri net. The focus is on the automatic construction of such a Petri net based on observed data. Systems are considered that are difficult to control due to the problems in modeling system dynamics as well as in applying mathematical tools to derive control laws. The approach is to automatically learn state transition laws based on a globally competitive and locally cooperative algorithm, to model the state transition laws by using a Petri net, and to automatically extract an optimal control law as well as to evaluate system controllability from the Petri net that is constructed