Gradient-free stochastic optimization algorithms are well-known for finding suitable parameter configurations over independent runs ubiquitously. Attaining low variability of convergence performance through independent runs is crucial to allow further generalization over distinct problem domains. This paper investigates the performance of a differential particle system in stabilizing a nonlinear inverted pendulum under diverse and challenging initial conditions. Compared to the relevant algorithms in the literature, our experiments show the feasibility of achieving lower convergence variability to stabilize a nonlinear pendulum over independent runs and initial conditions within a reasonable computational load.