This research addresses the problem of global path planning for a mobile robot in grid-based environments, where the path must simultaneously be short, collision-free, and dynamically smooth. The use of crisp fitness functions in such problems typically leads to severe discontinuities in the search space and high fluctuations in solution quality across repeated runs. The objective of this study is to improve stability and reduce the variance of the results of an improved version of the Bat Algorithm, without modifying its core update equations. To achieve this, a Mamdani-type fuzzy fitness system is designed. This system evaluates three linguistic inputs using continuous, overlapping triangular and trapezoidal membership functions. The fuzzy rule base is configured such that safety is prioritized over efficiency and smoothness, and its output is a normalized cost in the range [0,1] that replaces the crisp fitness value. The implementation is compared with a purely crisp version across multiple independent runs. Experimental results show that the average path length in the fuzzy case remains within an acceptable range, while the coefficient of variation of path length and fitness value decreases from approximately 0.38 to about 0.15 and 0.16, respectively, and the variance is reduced by approximately 90–95%. This significant reduction in dispersion indicates a smoother optimization landscape and more stable algorithmic convergence. The findings demonstrate that adding a fuzzy fitness layer on top of metaheuristics can substantially enhance the robustness and reliability of mobile robot path planning and can be readily generalized to other grid-based planning problems.