Research on AGV path tracking method based on global vision and reinforcement learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Automated Guided Vehicle (AGV) path following is critical for efficient logistics and manufacturing.
• Traditional AGV control algorithms face challenges in dynamic and complex environments.
• Developing robust path tracking control is essential for reliable AGV operation.

Purpose of the Study:

• To propose a novel AGV path tracking control method.
• To leverage global vision and reinforcement learning for enhanced AGV navigation.
• To improve the robustness and feasibility of AGV path following.

Main Methods:

• Utilizing visual sensors to capture global views and identify AGV and obstacle positions via target detection.
• Employing path planning algorithms to create a virtual environment for simulation.
• Integrating real-world AGV position and pose into the virtual environment for training.
• Inputting virtual camera images into a reinforcement learning model to generate control actions for the physical AGV.

Main Results:

• The proposed method successfully plans driving paths in various environments.
• The system demonstrates effective control of the AGV along specified paths.
• Experimental results validate the robustness and feasibility of the reinforcement learning-based approach.

Conclusions:

• The global vision and reinforcement learning-based AGV path tracking control method is effective.
• This approach offers strong robustness and feasibility for AGV navigation.
• The method provides a viable solution for complex AGV path following challenges.