The Randomized Potential Field Method works very well for most cases. However, our implementation of this method does run into trouble when it encounters a non-holonomic problem with highly restricted inputs, such as the forward-only car.
The problem arises because it can't choose random states, only random inputs. This severely limits the number of potential random states thus taking away the integrity of our method. Because PlannerDeltaT is fixed, we end up with only one state for each potential input. By varying PlannerDeltaT it may be possible to obtain a better random sampling. This could be implemented in any future versions of the RPF.
