Based on the exponential squared loss function, we propose a robust estimation method for the parametric parts in the partial linear models. By using the kernel approximation and transformation, we first absorb the nonparametric part. Then by using the exponential squared loss function to deduce the influence of outliers, we propose robust generalized estimation equations and empirical likelihood ratio function for estimation and inference. Under some regularity conditions, we show that the resulting estimators are consistent and asymptotic normality. Simulation results also illustrate the robustness and efficiency advantages of our method.