Visual Sensing and Depth Perception for Welding Robots and Their Industrial Applications

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision
• Manufacturing Technology

Background:

• Advancements in vision sensing, AI, and robotics necessitate improved sensors for intelligent welding manufacturing.
• Depth perception is a critical bottleneck hindering the development of advanced welding sensors.

Purpose of the Study:

• To review and assess active and passive sensing methods for depth perception in robotic welding.
• To explore deep learning applications for enhancing depth perception in robotic welding processes.
• To analyze the current state and future directions of visual perception for welding robots.

Main Methods:

• Classification and elaboration of depth perception mechanisms based on monocular, binocular, and multi-view vision.
• Exploration of deep learning principles for robotic welding depth perception.
• Analysis of 2662 articles, citing 152 references.

Main Results:

• Assessment of various depth perception methods, including active and passive sensing.
• Discussion on the application of deep learning in robotic welding visual perception across industrial scenarios.
• Identification of current challenges and proposed countermeasures in welding robot visual perception technology.

Conclusions:

• Future research should focus on deep learning for object detection, transfer learning for robot adaptation, and multi-modal sensor fusion.
• Integration of models and hardware, alongside expert collaboration, is crucial for designing effective multi-modal sensor fusion architectures.
• Addressing current limitations and exploring novel approaches will drive the evolution of intelligent welding manufacturing.