SUMMARY

Pneumatic actuation is frequently applied to situations that warrant inherent compliance, such as prostheses or walking robots, i.e., natural motions and applications in which interaction with humans/the environment are anticipated. However, compliance leads to control challenges that are commonly countered using aggressive controllers like sliding mode (SMC) or high-gain PID control, producing stiff system dynamics. Even hybrid force-position controller dynamics are ultimately subject to a clear trade-off of compliance and accuracy. In this thesis, this challenge is addressed via a constrained Model Predictive Controller that treats compliance as a bound rather than a target to achieve compliant tracking. A comprehensive literature review explores the state-of-the-art and defines performance targets, and a set of 1 degree of freedom (DoF) tests is established to compare controllers and convert controller aims into quantitative design specifications. Four benchmark controllers that span the stiffness-accuracy spectrum -- SMC, Linear Quadratic Tracking, PID, and Impedance Control -- are implemented in simulation and on hardware, and are used to produce baseline results and verify performance targets. The MPC is implemented with admittance and impedance constraints and compared to benchmarks on a 1-DoF system. New friction compensation methods are introduced that leverage the predictive structure to improve friction compensation for slow systems, and are compared to additive compensation methods. Results show that constrained MPC enforces impedance bounds on a tracked system, and achieve results with accuracy comparable to the best benchmark performance at a given compliance bound. Additionally, the highest tracking accuracy achieved with MPCs ultimately happens at the minimum necessitated impedance, without a-priori knowledge of that impedance bound. Results are shown to extend to a multi-DoF system using a planar robotic arm subject to unexpected disturbances.