UC Berkeley

Our society is witnessing an unprecedented, enduring, and pervasive aging process. With more and more people requiring walking assistance, the demand for gait physical therapy and rehabilitation has increased rapidly over the years. Current gait rehabilitation therapy relies on physical therapists to make evaluations and provide manual assistance, which is inaccurate and labor intensive. Introducing mechatronic techniques into gait rehabilitation can significantly improve the diagnosis and correction of gait abnormalities. This dissertation proposes a network-based gait rehabilitation system (NGRS) to achieve in-home rehabilitation, with physical therapists monitoring the rehabilitation training progress remotely.

The NGRS includes two main components: a wireless human motion monitoring system for gait analysis and a robotic assistive device controlled wirelessly for active gait correction. The NGRS is clearly a cyber-physical system (CPS) since it involves sensors and actuation devices, communication networks, and computational resources for perception and control. A general design framework is proposed for the design of CPS and the NGRS is developed using this framework as an example. This dissertation is divided into two parts, with the first part handling motion perception and gait analysis, and the second part developing fundamental networked control approaches for the robotic assistive device.

In the first part, design of the wireless human motion monitoring system is illustrated. The system includes a pair of wireless smart shoes embedded with barometric sensors for gait phase detection, and several wireless joint angle sensors for joint kinematic analysis. User interfaces are developed on both a laptop and an iPad to demonstrate processed sensory data as real-time visual feedback. It is shown that the sensory data make it more convenient and accurate to distinguish pathological gait from normal gait. Clinical effectiveness of visual feedback is verified with 24 stroke and Parkinson's disease patients.

In the second part, networked control systems (NCSs) are designed to achieve wireless motion control of the robotic assistive device. In this part, emphasis is given to the compensation of time delay and packet loss during wireless communication. For time delay compensation, two different approaches are presented. The first approach utilizes the delay measurement to build an equivalent system model, and an optimal preview controller is designed to utilize future reference signals for additional feedforward control. A double disturbance observer (DDOB) is developed to adaptively compensate for the time delay without any delay models or measurements. For packet loss compensation, two approaches are developed as well based on the Bernoulli packet loss model. The first technique is an extension of the modified LQG (MLQG) controller by adding a disturbance observer (DOB) for robustness enhancement. The preview control is also employed for packet loss compensation as the second approach. All techniques proposed in this part are validated by simulation and experimental results.