Abstract In response to the problems of local optimal convergence, insufficient path smoothness, and redundant inflection points in traditional ant colony algorithms in path planning, this paper focuses on algorithm optimization based on a double-layer ant colony system framework. Through systematic parameter sensitivity analysis experiments, determine the optimal initial configuration of core parameters such as pheromone volatility rate, and a two-layered Ant Colony Optimization mechanism (TLA-ACO) is proposed. This improved scheme adopts a hybrid search strategy of common ant colonies and elite ant colonies, constructs a dynamic pheromone update model, and significantly improves the global search ability and path quality of the algorithm. At the core mechanism design level, the research conducts multi-dimensional optimization of key parameters such as pheromone attenuation coefficient and path enhancement factor and establishes a nonlinear attenuation adjustment model. By integrating the angle penalty evaluation system and the potential energy guidance heuristic function, the smoothness and geometric rationality of the path topological structure can be effectively improved. To verify the robustness of parameter configuration, this study designed a gradient comparison experiment between pheromone weight coefficients and heuristic functions. All tests used a uniform random seed to ensure the reproducibility of the results. In the post-processing stage of the path, the collision detection redundancy elimination algorithm is innovatively combined with the quadratic Bézier curve fitting technique to construct a global path optimization framework. The results of MATLAB simulation experiments show that in a 30 × 30 grid environment, the TLA-ACO algorithm reduces the original path length by 7.24% and 3.84% compared to traditional ant colony algorithm and DL-ACO algorithm, respectively. After smoothing, the path length is further reduced by 11.78% and 6.85%; In a 40 × 40 complex scenario, the path optimization amplitudes reached 8.63% and 1.33% (original path) and 12.01% and 3.85% (smooth path), respectively, verifying the superior performance of the algorithm in complex topological environments. There are also varying degrees of optimization in the 50 × 50 grid map. And added real-world validation.
Yang et al. (Wed,) studied this question.