Active structural control framework using policy-gradient reinforcement learning