The on-line simplex search is an implementation of the well known Nelder-Mead optimization algorithm. The usual simplex search found in Matlab is not appropriate for an on-line implementation, because the state variables used are not persistent and the algorithm is not implemented to be reentrant. The implementation provided here is reentrant.
Thus, a sample vertex generated by the algorithm can be put out and be evaluated. After the system has settled to the evaluated vertex a new cost value can be read, after which the function can be reentered with the costs of previous vertices "remembered" due to persistent variables.