Surface roughness serves as a critical indicator of robotic machining quality for TC4 workpieces, significantly influencing functional performance and service life. However, existing approaches still struggle to achieve accurate real-time prediction of surface roughness during the robotic machining process. In order to address this gap, a novel real-time prediction model based on multisensor signal fusion using the PSO-XGBoost algorithm is developed in the paper. Initially, acoustic emission (AE) signals and triaxial vibration signals are collected and decomposed using Ensemble Empirical Mode Decomposition (EEMD) and Wavelet Packet Transform (WPT) during the robotic grinding process. Subsequently, an optimized PSO-XGBoost prediction model is proposed to enhance the generalization capability and prediction accuracy. Finally, experimental validation is performed on a robotic grinding platform for TC4 workpieces under varying machining parameters to achieve a stable and accurate real-time prediction of surface roughness compared to conventional methods.