This paper addresses the limitations of traditional ant colony optimization (ACO) in path planning, such as slow convergence, susceptibility to local optima, and numerous path inflection points, by proposing an improved ACO algorithm. This algorithm integrates a heterogeneous adaptive mechanism, angle penalty, and an elite strategy, and systematically verifies its generalization performance. By constructing a heterogeneous-homogeneous dual-population collaborative architecture, combining two ant populations with different characteristics, the algorithm′s global search capability in various environments is enhanced, effectively avoiding premature convergence. Introducing a direction-aware angle penalty factor avoids unnecessary path inflection points, optimizes path smoothness, and improves the algorithm′s adaptability to complex terrain by adding angle penalties to path planning. The elite-weighted pheromone update strategy allows the influence of excellent solutions to be more fully reflected in the pheromone update process, accelerating the convergence process and improving stability. In the comparative experiment of multi-scale grid maps, the algorithm proposed in this paper showed excellent generalization performance and robustness: in a complex 50×50 environment, compared with the traditional ACO algorithm, the path length was reduced by 14.1% and the inflection point was reduced by 69.4%; compared to existing improved algorithms, the path length was shortened by 8.4%, the inflection point was reduced by 66.6%, and the number of iterations was reduced by 82.6%. The real vehicle experiment of the automated guided vehicle (AGV) further verified the generalization ability of the algorithm in the real scene, the path length was shortened by 11.1%, and the inflection point was reduced by 78.2%. This study innovatively proposed a population heterogeneous adaptive scheduling mechanism, a direction-aware angle penalty strategy, and an elite pheromone weighted update method, which significantly improved the generalization performance of the ACO algorithm and provided reliable technical support for the practical application of mobile robot navigation systems.