Reinforcement Learning Enhanced Archimedes Optimization Algorithm for Robot Path Planning
Muhammad Shafiqul Islam, Mohd Ashraf Ahmad, Julakha Jahan Jui · 2025
This paper introduces the Reinforcement Learning Enhanced Archimedes Optimization Algorithm (RLAOA), an improved variant of the traditional Archimedes Optimization Algorithm (AOA) that incorporates reinforcement learning via Q-learning to enhance performance. The proposed RL-AOA leverages Q-learning to dynamically balance exploration and exploitation, leading to more efficient optimization. To evaluate its effectiveness, RL-AOA has been applied to a path planning problem in a 2D environment with obstacles. The algorithm optimizes a trajectory represented by control points interpolated with cubic splines while minimizing a cost function that accounts for both path length and obstacle avoidance. Simulation results demonstrate that RL-AOA significantly outperforms the original AOA as well as other metaheuristic algorithms, namely SCA and ABC, in terms of convergence speed, mean cost, and best-case cost. Moreover, RL-AOA exhibits greater stability with the smallest standard deviation. Convergence analysis further confirms that RL-AOA achieves superior solutions in fewer iterations, validating its potential for complex path planning tasks.