In this thesis we investigate control across stochastic drop-out channels. We seek optimal linear controllers for mean-square stability that make use of the knowledge of whether a packet is received. We establish a fundamental bound on drop-out probability allowable for stabilization, which in some cases is tight. When tight, a convex optimization provides controller design. The main result is a remote stabilization technique that always achieves this bound via acknowledgement from the actuation receiver. Controller information structure and decentralization issues are considered. The theory is then applied to the inverted pendulum experiment Pendubot. The Pendubot is a nonlinear plant that balances the links of an inverted pendulum via a control torque and optical sensors for position. A control strategy is developed using the stabilization technique derived above together with linearization and discretization of the apparatus. The complete design procedure is documented leading to a successful controller. Included are the encountered hardware issues, software issues in Matlab and C programming, theoretical issues, and experiment results. A mock drop-out network is simulated via C programming. The experiments validate that the theoretical design technique actually works, and that the theoretical bounds on allowable drop-out have significant practical bearing.