UC Berkeley

As the number of patients requiring gait rehabilitation treatments is increasing, assistive systems for gait rehabilitation are being actively investigated. Assistive systems enable more efficient rehabilitation by providing objective values for indicating the patient's status and assistive torque for practicing normal trajectories for rehabilitation. This thesis investigates several mechatronic technologies of assistive systems for gait rehabilitation, including (1) estimation and evaluation of the patient's status, (2) monitoring systems, (3) control of assistive systems, and (4) implementation of rehabilitation algorithms.

Estimation and evaluation of a patient's status based on pertinent measurements is the first step toward determining appropriate rehabilitation intervention methods. This thesis introduces an algorithm that estimates gait phases using a hidden Markov model (HMM) based on ground reaction forces (GRFs) measured by force sensors embedded in shoes, called Smart Shoes. The GRFs and the center of the GRFs (CoGRF) are used for observing the patient's status, and gait abnormality is calculated based on deviations from healthy GRF levels. This information is supplied to the monitoring system, which is implemented as a mobile system and a tele-system using the Internet. Assistive torque is required for seriously impaired patients to achieve the desired motion or practice normal trajectories. Ideal force mode control is necessary for natural interactions between the assistive system and the patient. In this thesis, robust control algorithms for precise and safe generation of the desired torque are discussed. The proposed algorithms have been applied to the previously developed assistive systems such as a rotary series elastic actuator (RSEA), a compact rotary series elastic actuator (cRSEA), and a cable-driven assistive system. As a decision-making process for rehabilitation, both a power augmentation method and a rehabilitation method are discussed. For the power augmentation method, the joint torque of the lower extremities is estimated using a human model with seven links and four different ground contact conditions. For gait rehabilitation, a potential field around the desired trajectory and an iterative learning algorithm inspired by repetitive gait motions are proposed for determining the desired assistive torque. The proposed methods have been verified experimentally, including clinical tests using actual patients.