This paper presents a novel spatio-temporal long short-term memory (ST-LSTM) network for 3D human action recognition. The authors extend traditional LSTM to both spatial and temporal domains to model the contextual dependencies in the input data. They propose a tree-structure-based traversal method to better capture the spatial dependencies between joints, as the simple chain model often ignores the kinematic relationships. Additionally, a "trust gate" mechanism is introduced to handle noisy and occluded 3D skeleton data by assessing the reliability of input data at each spatio-temporal step. The proposed method achieves state-of-the-art performance on four challenging benchmark datasets, demonstrating its effectiveness in 3D human action recognition.This paper presents a novel spatio-temporal long short-term memory (ST-LSTM) network for 3D human action recognition. The authors extend traditional LSTM to both spatial and temporal domains to model the contextual dependencies in the input data. They propose a tree-structure-based traversal method to better capture the spatial dependencies between joints, as the simple chain model often ignores the kinematic relationships. Additionally, a "trust gate" mechanism is introduced to handle noisy and occluded 3D skeleton data by assessing the reliability of input data at each spatio-temporal step. The proposed method achieves state-of-the-art performance on four challenging benchmark datasets, demonstrating its effectiveness in 3D human action recognition.