Research on Deep Learning-Based Intelligent Kiln Car Unstacking Robot Vision Recognition and Path Planning

With the advancement of Industry 4.0, intelligent manufacturing has imposed higher requirements for automated production. Kiln car unstacking, as a critical process in the ceramics and building materials industries, suffers from low efficiency and high safety risks when performed manually. This paper presents an intelligent kiln car unstacking robot system based on deep learning that integrates vision recognition and path planning technologies. The system employs an improved YOLOv8 algorithm for product detection and localization, enhancing recognition accuracy through attention mechanisms and multi-scale feature fusion techniques. For path planning, an intelligent planner based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is designed, incorporating prioritized experience replay and LSTM technology to achieve efficient navigation in dynamic environments. The system adopts a hierarchical architecture comprising perception, decision-making, and execution layers, ensuring real-time performance and reliability through multi-sensor fusion. Experimental results demonstrate that the improved YOLOv8 algorithm achieves 95.2% detection accuracy, representing an 8.7% improvement over the baseline. The TD3 path planning algorithm achieves a 96.8% success rate with 12.3% shorter path lengths and 45.6% reduced planning time. In practical industrial testing, the system improves work efficiency by 73.5% and reduces error rates by 89.2% compared to manual operations, validating the effectiveness of deep learning technology in complex industrial environments and providing important technical reference for intelligent manufacturing applications.