Design, Control, and Implementation of a Robotic Gait Rehabilitation System for Overground Gait Training

Robotic devices for gait rehabilitation have the potential to improve patient and caregiver safety, reduce therapy costs, and allow a larger number of patients to get access to physical therapy. Additionally, data collected from the robot's sensors may be used to assess impairment severity and track patient progress. However these devices also suffer from many drawbacks such as high cost, complexity, limited training capabilities, and constrained joint motions and postural responses. With these limitations in mind, this thesis introduces GaitEnable, a simply designed robotic gait trainer that combines an intelligent reactive controller, an actuated omnidirectional mobile base and a passive body weight support system. In addition to describing the device and its control system, this thesis also presents results from a series of validation experiments performed to characterize the performance of the device. The results demonstrate that GaitEnable's control system ensures stable human-robot interactions, and that GaitEnable can assist and perturb a user's gait in a systematic manner. The experiments also confirm that GaitEnable's actuated omni-directional mobile allows users to walk more naturally as it reduces the motion constraints that the device imposes. In addition to describing the GaitEnable system, this thesis also focuses on the more general problem of interaction stability in coupled human-robot systems. Two novel trajectory manipulations for ensuring stable, oscillation-free interactions in admittance-controlled haptic devices are proposed. A Lyapunov stability analysis is used to show that the proposed manipulations are stable in the sense of uniform ultimate boundedness. Also, an extensive set of experiments confirm that the impedance manipulations allow the display of large apparent inertia reductions, ensure stable interactions, and are robust to actuator saturation and model uncertainties. These features make them ideal for use with robotic systems that are attached to humans (e.g., the GaitEnable system). Use of these manipulations is also extended to typical position control problems, and additional experimental results confirm the manipulations help eliminate chatter and provide a better transient response and steady-state tracking error.