Utilization of automatic differentiation in control algorithms

The authors compare an automatic differentiation code and a symbolic approach for the problem of integrating a nonlinear prescribed path control problem. For a small problem either approach is suitable; for problems of moderate size, where many evaluations of the differentiated quantities are required, or for problems where the symbolic computation can be done before a simulation, a symbolically generated FORTRAN subroutine can be preferable. The symbolic approach is also somewhat more natural and a little easier to understand at first. For larger problems, or modest problems involving a moderate number of derivative evaluations, the automatic differentiation code was superior. Automatic differentiation is also superior whenever memory utilization becomes a factor with the symbolic approach