KD-Former: Kinematic and dynamic coupled transformer network for 3D human motion prediction

•We propose a novel non-autoregressive Kinematic and Dynamic coupled transFormer (KD-Former) network for 3D human motion prediction.•Our KD-Former leverages the complementary characteristics of motion kinematics and dynamics for performance improvements.•We formulate a simplified reduced-order dynamic algorithm, largely enhancing computation efficiency and prediction error. Extensive experiments on Human 3.6M and CMU MoCap datasets demonstrate the superior performance of our method.