Autonomous learning for control systems with continuous state/action space