Create robust standard errors in a garch model

Hi, I am estimating a garch model and I would like to obtain robust standard errors and t-statistics after the estimation. To estimate I have created a negative log likelihood and I used fmincon. Then I used the hessian created by fmincon to obtain the standard errors (I divide it by the sample size,I take the invers, the diagonal and at the end the square root). To obtain the robust standard errors I want to try sqrt(diag(inv(hessian)*(grad*grad')*(inv(hessian)'))/(2*T_n)) but I am not sure this is the right way to proceed. Can someone give me a hint about it?