Prediction of End-Effector Motion in Industrial Robots using LSTM Neural Networks: A Deep Learning Approach in Robot Dynamics

Industrial robots perform various tasks like welding, drilling or assembly in manufacturing industry. For improved precision, safety, and task efficiency in automated manufacturing, the prediction of the position of the end-effector is important, since an unknown or inaccurate position can lead to misalignment, defective products, equipment collisions, or even pose safety risks to the factory worker. Traditional physics-based and simple models have limited predictive accuracy when it comes to nonlinear robot behavior and external disturbances and diverse motion patterns such as linear, curved and circular trajectories. Due to this, a deep learning framework based on long-short-term memory (LSTM) networks is proposed in this paper to predict the position of industrial robots end-effector. In this deep learning model, temporal dependencies on sequential sensor data were learned and short-term motion was predicted with high accuracy. The predicted end-effector position was then evaluated on the new data and compared with the ground truth to determine the generalization power of the prediction model. The proposed models can be used in future work to support predictive trajectory planning, early anomaly detection, and real-time motion adaptation.