Robust and Adaptive Optimal Control Methods for a Hybrid Neuroprosthesis

Functional electrical stimulation (FES) is an external application of electrical pulses to skeletal muscles to produce desired limb movements. It is prescribed as a rehabilitation intervention to restore standing and walking functions in people with paraplegia. However, its clinical implementation is hindered by a rapid onset of muscle fatigue that limits its use for longer durations. To overcome the FES-induced muscle fatigue, hybrid neuroprostheses that combine FES with powered exoskeletons were proposed recently. However, how to coordinate FES and powered exoskeleton in a hybrid neuroprosthesis still remains an open issue. The long-term goal of this research is to develop control methods that can optimally coordinate FES and the powered exoskeleton by considering muscle fatigue dynamics during standing and walking activities. The research objective in this dissertation was to derive robust and adaptive optimal control methods for two hybrid neuroprostheses: a hybrid leg extension machine (HLEM) and a full lower-body neuroprosthesis (FLBN). Firstly, a model predictive control (MPC) method that coordinates FES and an electric motor in the HLEM is developed. However, due to inaccurate system identification, day-today variations in the model, and partially measurable state, it is challenging to implement this method in a clinical setting. Therefore, robust and adaptive versions of the MPC method were derived. To overcome modeling uncertainties, a tube-based robust MPC was derived. This MPC has a feedback controller that can drive the actual state into a region centered by the nominal state. This ensures recursive feasibility and stability despite disturbances. Later, a recurrent neural network (RNN) was developed to capture the non-autonomous behavior in the musculoskeletal system, and then a nonlinear MPC and a reinforcement learning (RL) method were derived to sub-optimally compute the control actions for the system. To achieve a standing-up motion, a ratio-allocation method was developed to determine the ratio of the FES-induced torque to the motor torque at the knee joint. The dynamically varied estimated muscle fatigue was used as an index that guided the optimal allocation. Experiments were performed to validate the robust and adaptive methods. The results show a potential of the proposed methods for clinical implementation.