<p>Structural vibrations and resonance induced by pose variations during high-speed motion of high-precision assembly robots significantly amplify positioning errors. Effective prediction of dynamic properties under varying poses is critical to mitigate risks and optimize performance. This study proposes an improved method for rapid and accurate prediction of FRFs at arbitrary poses within the robot’s workspace, enabling subsequent identification of modal parameters to support vibration suppression and trajectory optimization. An enhanced inverse distance weighting model is developed, whose core innovation lies in replacing Euclidean distance with Mahalanobis distance to establish a more precise mapping from pose space to FRFs. Modal parameters are then extracted from predicted FRF curves via frequency-domain parameter identification. Experimental results demonstrate that the proposed model efficiently and accurately predicts FRF curves across arbitrary poses. Modal identification reveals that the average prediction accuracy for natural frequencies is significantly improved by approximately 23 % compared to conventional methods. The Mahalanobis distance-based improved model effectively addresses pose-dependent dynamic prediction challenges, substantially enhancing the accuracy of key modal parameters and providing a reliable technical foundation for enhancing performance in high-speed, high-precision assembly robots.</p>