Virtual, Augmented, and Mixed Reality Robotics-Assisted Deep Reinforcement Learning Towards Smart Manufacturing

Area of Science:

• Robotics and Automation
• Manufacturing Engineering
• Computer Science

Background:

• Welding robots are vital for precision and efficiency in modern manufacturing.
• Accurate simulation is crucial for optimizing welding robot performance and minimizing errors.
• Existing simulation methods require enhancement for greater realism and fidelity.

Purpose of the Study:

• To present a novel approach for enhancing welding robot simulations.
• To improve the fidelity and realism of robot behavior simulations.
• To accelerate the learning and optimization of welding robot operations using advanced techniques.

Main Methods:

• Integration of the Virtual, Augmented, and Mixed Reality (VAM) simulation platform with existing techniques.
• Application of deep reinforcement learning (DRL) for task offloading and real-time trajectory planning.
• Utilizing Adaptive Neuro-Fuzzy Inference System (ANFIS) for control strategy comparison.

Main Results:

• The VAM platform provides a dynamic and realistic environment for simulating robot actions and interactions.
• DRL integration enhances simulation accuracy and improves real-time decision-making for robots.
• Experimental results indicate ANFIS outperforms traditional PID and Fuzzy Logic Control (FLC) in accuracy and convergence speed.

Conclusions:

• Combining advanced simulation platforms like VAM with machine learning, specifically DRL, significantly advances industrial robot capabilities.
• The proposed approach enhances welding robot simulation realism and operational efficiency in smart manufacturing.
• ANFIS demonstrates superior performance over PID and FLC for welding robot control.