Chunling Wang
Abstract Industrial robot systems require high accuracy, adaptability, and reliability to carry out complex operations such as assembly, inspection, pick-and-place, and welding. Traditional control methods, i.e., PID, PCA, and HMM-based methods, are likely to be constrained in handling dynamic disturbances, sensor noise, and single-modality imprecision, resulting in less-than-optimal accuracy and reliability. To address these issues, this paper presents a CNN + LSTM-based Hybrid Control Module (HCM) integrating multimodal perception fusion from joint encoders, 3D vision systems, and laser trackers. Experimental results show excellent performance, with accuracy = 0.9823, precision = 0.9820, recall = 0.9819, F1-score = 0.9821, FNR = 0.018, FPR = 0.004, and ROC-AUC values over 0.994. The method combines hierarchical sensor fusion, integrating high-frequency rapid signals with low-frequency precise measurements, enabling real-time error compensation and tolerant control under various operating conditions. The originality of this work is that it achieves sub-millimetre positioning accuracy without sacrificing high precision and recall across a range of industrial tasks. Comparative analysis shows that the proposed HCM outperforms PCA (accuracy 83%) and HMM (accuracy 87.2%) and performs comparably to CNN-BiLSTM (precision 0.83) and YOLOv10 (precision 0.998), confirming its superior generalization and reliability. These findings suggest that hybrid control, driven by multimodal perception, can deliver effective precision, stability, and flexibility in industrial robotic processes. This method offers a stable and scalable option for future smart manufacturing systems, demonstrating high operating reliability and overcoming the variability of dynamic environments and sensor limitations.