Heuristic Search Landscape
Make use of simple knowledge of problem domain to provide landscape helpful to search
Notes:
This simple knowledge of our problem domain provides a “landscape” helpful to search. Here we see one such landscape. In the test problem shown, we’re given bounds on possible parameters for viscous friction and load inertia and all other parameters are fixed. The search landscape slopes downward towards regions where the trajectories are closer to stalling. A function value of zero indicates a stall.
The problem shown was actually sought and chosen for its difficulty. For most stepper motor verification problems encountered, simple local optimization of generally simple landscapes rapidly led to stall scenarios (if they existed). This problem however illustrates the need for good global optimization techniques: There are local minima along the edges of this space, and while the landscape does indeed slope downward towards stalling scenarios in the entire parameter space, the landscape largely slopes away from the rare stall scenarios shown in the lower corner of our bounded space.