Multi-robot path planning with obstacle avoidance using hybrid optimization and deep learning approaches

Mobile robot navigation relies heavily on path planning to determine optimal routes and avoid obstacles as they operate within dynamic surroundings. However, many current obstacle avoidance methods remain unstable and lack practicality, limiting their effectiveness in real‑world applications. Therefore, this research presents a deep learning (DL)-based path planning method for mobile robots, primarily aimed at improving obstacle avoidance. The primary novelty of the proposed framework lies in two key contributions. First, multi-objective path planning is addressed through a new Fractional Mayfly Deer Hunting Optimization (FMDHO) algorithm, in which Fractional Calculus (FC), the Mayfly Algorithm (MA), and the Deer Hunting Optimization Algorithm (DHOA) are incorporated into a unified optimization framework. This enables the resulting paths to be shorter, smoother, and free from collisions. Second, a Le-Q Net model is proposed by combining LeNet with a Deep Q-Learning Network (DQN) to enhance multi-robot path planning and obstacle avoidance in complex environments. The framework begins with simulating the workspace dimensions and obstacles, and then the obstacle space is expanded for robot safety. The subsequent stage focuses on generating safe paths without collisions, where FMDHO is employed to optimize the paths according to multiple objectives. In the final stage, Le-Q Net is utilized to determine the paths for multiple robots. As a result, the proposed FMDHO achieved Path smoothness of 0.758, Fitness of 0.296, Path length of 1329.42 m, and arrival time of 0.626 s. Similarly, the Le‑Q Net model obtained path smoothness of 0.843, fitness of 0.309, path length of 1403.91 m, and arrival time of 0.621 s, under fixed obstacles and multiple targets. Not applicable. This study does not involve a clinical trial.