RL-QPSO net: deep reinforcement learning-enhanced QPSO for efficient mobile robot path planning.

Area of Science:

• Robotics
• Artificial Intelligence
• Computational Science

Background:

• Traditional path planning methods struggle with complex, dynamic environments, often getting stuck in local optima and lacking real-time efficiency.
• Existing algorithms like genetic algorithms, Dijkstra's, and Floyd's have high computational costs and are unsuitable for dynamic settings.

Purpose of the Study:

• To introduce a new path planning model, RL-QPSO Net, for mobile robots.
• To enhance global optimality and adaptability in path planning using a hybrid approach.

Main Methods:

• Developed RL-QPSO Net, integrating Quantum-behaved Particle Swarm Optimization (QPSO) for global optimization and Deep Reinforcement Learning (DRL) for real-time adaptation.
• Employed a dual control mechanism for path optimization and environmental adaptation.

Main Results:

• RL-QPSO Net demonstrated superior performance over traditional methods in accuracy and computational efficiency across multiple datasets (Cityscapes, NYU Depth V2, Mapillary Vistas, ApolloScape).
• The model achieved significant improvements, offering an effective solution for real-time mobile robot path planning.

Conclusions:

• The RL-QPSO Net provides an effective and efficient path planning solution for mobile robots in complex, dynamic environments.
• Future work may extend this method to resource-limited environments for broader practical applications.