The industrial robot industry in the Pearl River Delta region is undergoing a transformation towards intelligent manufacturing, facing an urgent need to improve production efficiency and operational accuracy. To enhance the recognition and adaptation capabilities of robots in complex environments, research has been conducted to develop a robot motion recognition algorithm based on an LSTM-RNN model. This aims to overcome the vanishing and exploding gradient problems faced by traditional RNNs when processing data with long-term dependencies. Experimental results demonstrate that the ILSTM-RNN model performs exceptionally well in robot motion recognition tasks, achieving an accuracy rate of 93.2% after 1000 iterations, which is significantly better than traditional RNN (75.3%) and LSTM (85.0%) models. The loss value of the model eventually drops to 0.43, indicating high training stability and convergence speed. Additionally, ILSTM-RNN exhibits outstanding performance in industrial robot applications, with a 21.3% increase in production efficiency, a 19.7% improvement in quality conformance rate, and a 26.7% reduction in response time. By optimizing the algorithm structure and regularization strategies, the ILSTM-RNN model significantly improves the accuracy and training stability of robot motion recognition and demonstrates substantial potential in practical industrial applications, particularly suitable for tasks requiring high real-time performance and precision.