1: \begin{abstract}

2: We consider the static output feedback control for Linear Quadratic Regulator problems with structured constraints

3: under the assumption that system parameters are unknown.

4: To solve the problem in the model free setting, we propose the policy gradient algorithm based on the gradient projection method

5: and show its global convergence to $\varepsilon$-stationary points.

6: In addition, we introduce a variance reduction technique

7: and show both theoretically and numerically

8: that it significantly reduces the variance in the gradient estimation.

9: We also show in the numerical experiments that the model free approach efficiently solves the problem.

10: \end{abstract}

11: