Abstract: The artificial fish swarm algorithm is easy to fall into the local optimum for robot global path planning. A hybrid improved Artificial Fish Swarm Algorithm (HIAFSA) is proposed. Firstly, the sub-optimal path is determined by A* algorithm, and then the adaptive behavior of artificial fish swarm algorithm is improved based on the inertia weight factor, and the attenuation function $\alpha$ is introduced to improve the visual range and moving step length of the artificial fish, balance the global path planning and local path planning, and further improve the convergence speed and quality of the solution. The experimental results show that the hybrid improved artificial fish swarm algorithm has been improved in avoiding local optimum, convergence speed and precision.
Introduction
In autonomous robotics, efficient path planning—finding the shortest collision-free route from a start point to a target—is a fundamental challenge. The Artificial Fish Swarm Algorithm (AFSA) is a popular bio-inspired metaheuristic that mimics the social behaviors of fish (foraging, swarming, and following) to search for optimal solutions. While robust, standard AFSA often suffers from slow convergence and a tendency to get trapped in local optima, especially in complex environments. Furthermore, its fixed parameters for visual range and step length limit its ability to balance global exploration with local exploitation.
Methodology: The HIAFSA Approach
To address these limitations, the researchers developed the Hybrid Improved Artificial Fish Swarm Algorithm (HIAFSA). This method integrates deterministic search with stochastic swarm intelligence through three primary enhancements:
1. A* Initialization
The algorithm begins by using the A* algorithm to find an initial sub-optimal path. This preliminary step provides a skeleton for the path, which is then used to initialize the artificial fish population. By "priming" the swarm with a rough idea of the route, the search space is significantly reduced, preventing the fish from wasting computational resources in irrelevant areas of the map.
2. Dynamic Attenuation Function ($\alpha$)
In standard AFSA, the visual range ($Visual$) and step length ($Step$) are constant. HIAFSA introduces an attenuation function, $\alpha$, which dynamically adjusts these parameters as the iteration count ($t$) increases.
As shown in the figure above, the parameter $\alpha$ decays over time. Early in the search, a high $\alpha$ allows fish to have a wide visual range and large steps for global exploration. As the algorithm converges, $\alpha$ decreases, making the fish "near-sighted" with smaller steps to perform fine-tuned local optimization and refine the path's precision.
3. Nonlinear Inertia Weight ($\omega$)
Drawing inspiration from Particle Swarm Optimization (PSO), a nonlinear inertia weight factor is added to the fish's movement equations. This weight helps the algorithm maintain momentum from previous movements, further smoothing the transition from broad searching to local refinement.
Experimental Setup
The performance of HIAFSA was evaluated using two distinct 40x40 grid maps representing different environmental complexities.
- Map 1 (Complex): Contains numerous, unevenly distributed obstacles to test the algorithm's ability to navigate tight spaces without getting stuck.
- Map 2 (Simple): Features a large central U-shaped obstacle, a classic "trap" for many local-search algorithms.
Key Results and Discussion
The HIAFSA was benchmarked against the standard AFSA and an improved version (IAFSA) over 50 experimental trials.
- Path Efficiency: In the complex environment (Map 1), HIAFSA achieved an average path length of approximately 49.0, a significant improvement over the standard AFSA's 66.4.
- Reliability: HIAFSA demonstrated superior stability, with a variance of only 0.0105 in Map 1, compared to 0.2437 for the standard version. This indicates that the hybrid approach consistently finds near-optimal paths.
- Convergence Speed: Iteration curves revealed that the combination of A* initialization and adaptive parameters allows HIAFSA to reach its optimal value much faster than competing methods.
Conclusion
By combining the deterministic strengths of the A* algorithm with the adaptive, stochastic nature of an improved Artificial Fish Swarm Algorithm, HIAFSA effectively overcomes the common pitfalls of traditional swarm-based path planning. The introduction of the dynamic attenuation function and inertia weights ensures a robust balance between global search and local precision. Future research will focus on extending this model to dynamic 3D environments and optimizing its computational efficiency for real-time deployment on embedded robotic systems.
Summary updated