Vision-Guided Robotic Assembly Using Deep Neural Networks

Abstract

Intelligent manufacturing is transforming traditional robotic assembly into adaptive, autonomous, and data-driven production systems. Conventional robotic assembly relies on pre-programmed trajectories, structured environments, and rule-based vision systems, limiting flexibility in handling complex components and uncertain operating conditions. This paper proposes a deep neural network (DNN)-based vision-guided robotic assembly framework that integrates computer vision, intelligent decision-making, and real-time robotic control. The framework consists of four modules: vision acquisition, deep feature learning, assembly intelligence, and robotic execution. Industrial cameras and depth sensors capture visual data, while convolutional neural networks (CNNs) perform object recognition and pose estimation. Extracted visual features are combined with motion planning to generate optimized assembly trajectories. Mathematical models for feature extraction, neural network optimization, and robotic coordinate transformation enhance system accuracy and reliability. Performance is evaluated using recognition accuracy, assembly precision, processing speed, adaptability, and operational efficiency. Experimental results demonstrate that the proposed framework outperforms conventional image processing and machine learning approaches, enabling accurate assembly of irregular components under uncertain conditions. The proposed approach enhances perception, autonomous decision-making, and intelligent adaptation, supporting flexible automation and next-generation smart manufacturing in Industry 4.0 environments.