A robotic exoskeleton is a powered wearable device that is capable of actuating one or more of a person's joints to physically interact with the user in a significant and controlled manner. The work presented in this thesis is the result of a collaboration with NASA's Dexterous Robotics Laboratory (DRL) to design a soft, wearable exoskeleton for neuro-rehabilitation, an application that requires accurate sensing and control of the interaction forces and torques between the robot and the human to be successfully implemented.

To achieve accurate interaction control, a well-established method of cascaded control with an inner velocity loop is applied to a series elastic actuator (SEA) on an experimental test bed. The test bed is made modular and reconfigurable to closely study the effect of the location of a custom-designed elastic force sensor placed within the conduit transmission. Proper implementation of the control scheme is verified through system identification techniques with comparison to model predictions.

Having the force sensor as close to the user interaction point as possible predictably results in a more accurate rendering of a virtual impedance to the user, despite the no load motion torque controller showing issues of steady-state error when the sensor is in this location. When the force sensor is located close to the stationary DC motor, a choice that is attractive for designers of the human interface to a soft exoskeleton, the degradation in the device's ability to render virtual impedance can be characterized by a model of the nonlinear friction interaction between the cable and the conduit transmission.