Pneumatic actuators are reliable and durable. They are also friendly to the environment, and low-cost compared with hydraulic systems and electrical actuation systems. One of the bottlenecks constraining the development of pneumatic systems is they suffer from low system energy efficiency due to misuse of compressed air, leakage, and the throttling losses of pneumatic components. We hypothesize that the energy efficiency of a position controlled pneumatic actuator can be improved by designing a better pneumatic circuit and controlling it with an advanced control algorithm. In this thesis, we study the position control of two different pneumatic circuits, known as direct pump control (DPC) and valve control (VC), and compare their position control performance and energy consumptions. The DPC circuit consists of a double acting pneumatic cylinder that is controlled by two positive displacement air pumps, and four ON/OFF solenoid valves. The VC circuit consists of the same cylinder controlled by four ON/OFF solenoid valves, and an air tank that is pressurized by a single pump. Nonlinear system models of the two circuits were developed, in which a novel air pump subsystem model and a novel friction model were proposed and validated by comparing simulation and experiment results. Discrete-valued model predictive control algorithms were then developed for each circuit. These algorithms were designed to find the optimal combination of pump and valves binary states for controlling the position of the load and reducing energy consumption. After tuning their parameters based on a series of short experiments, long duration experiments were conducted to allow the position control performance and energy consumption of the two circuits to be compared fairly. From the results of the long duration experiments, the root mean square error and mean absolute error of DPC are 19.77% and 13.42% lower than those of VC. However, the mean steady state error is 0.74 mm, and the mean overshoot is 1.73 mm, which are 17.46% and 53.10% higher than VC’s results. This means VC has better steady state precision, while DPC is superior at tracking the setpoint trajectory transients. Regarding the energy consumption, 92.84% of the system energy was saved using DPC with one working cycle, and 36.64% was saved when the number of working cycles was increased to 29, which proves DPC is more energy efficient.