This thesis presents the design and experimental application of the Equilibrium Point Hypothesis as a controller model for a programmable mechanical compliant manipulator. A planar manipulator was designed and constructed with two joints, each powered by a pair of antagonistic McKibben actuators (air muscles). Programmable mechanical compliant manipulators provide increased intrinsic safety and the ability to implement a controller based on the EP Hypothesis becomes possible. The EP Hypothesis presents a model describing how human arm motions may be controlled. A previously developed geometrically derived force model for air muscles was modified leading to the formulation of a linearizing and decoupling compensator. This compensator, in conjunction with a proportional, integral controller operating on air supplied to the muscles, provided stable control of the stiffness and EP of each joint of the manipulator. A benefit of this combined EP and stiffness control is that a single control strategy can be used both to control the manipulator position in free-space and to provide interaction control for contact tasks.

A series of experiments were performed to demonstrate the controller behaviour in free space, in transition from free space to contact, and in contact with the environment. The free space experiments were done mainly to characterize the controller behaviour. The transition task involves moving in free space to contacting a surface at different velocities and contact angles. The contact task is a wiping motion along a surface with a prescribed normal force. The effect on introducing an unexpected "bump" along the surface was examined, as were velocity effects.

The stable behavior during transition from free-space to contact is a notable result. Because the manipulator follows an equilibrium-point trajectory with a programmed stiffness, no additional compensation is required when contacting objects in the workspace. Additionally the precise location of the object is not important as the mechanical compliance of the manipulator compensates for small contact position errors.

The results of the surface wiping tasks showed that it is possible to generate a wiping EP and stiffness trajectory that results in the predicted normal force while wiping a surface. Additionally, the mechanical compliance of the manipulator allows for stable response to unpredicted disturbances such as the presence of a significant bump on the smooth surface.