Linear robust control techniques such as μ-synthesis can be used to design controllers for linear systems to guarantee specified performance criteria in the presence of modeling uncertainties, disturbances, and sensor noise. However, these techniques are rather uncommon in robotics due to the nonlinear nature of the plant where direct application would require large model uncertainties and therefore may only create a satisfactory controller if using lenient performance criteria. The inclusion of feedback linearization can rectify this by effectively converting the plant from a nonlinear system to a linear one, resulting in smaller model uncertainties.
This paper proposes the use of feedback linearization to enable the use of linear robust control techniques on nonlinear systems. This approach is applied to a provisional version of a powered pediatric lower-limb orthosis. Sine sweep experiments are conducted to determine frequency response data for the system with and without feedback linearization. Models are identified to match the recorded data using optimization for both cases. Uncertainties are manually applied such that they encapsulate the observed measurements. The amount of uncertainties in the two models are quantified and a comparison shows that the uncertainties in the feedback-linearized system are smaller than in the system without feedback linearization.