UC Berkeley

The need for intelligent assistive devices is growing. Due to advances in medicine, people are living longer and able to recover from severe neurological incidents, resulting in an increased population with neuromuscular weakness. In workplaces such as assembly lines, there is a high possibility of work-related fatigue or injury, such as when workers squat down or lift their arms during their work tasks. Assistive devices could help remedy loss of strength on their extremities as well as keep the work environment safe and productive, allowing these growing segments of the population in need of the devices to live more self-sufficient, productive, and higher-quality lives.

In the design of assistive systems, an important design goal is prolonged operational time, which requires the minimum usage of energy. Energy consumption can be reduced by modifying the mechanical characteristics of assistive systems according to the dynamic characteristics of the human body, which vary considerably between tasks. This dissertation investigates 1) the design of actuators with adjustable mechanical impedance, 2) control strategies to search for, and adjust to, a suitable mechanical impedance for assistance and 3) sensing technologies for classifying the tasks in which the human engages.

The first part of this dissertation characterizes a pneumatic variable stiffness actuator named an Active/Passive Pneumatic Actuator (AP2A). The actuator consists of an air cylinder and an array of solenoid valves. These valves and the corresponding switching algorithms tune the chamber pressures and make the AP2A function as a mechanical spring with desired stiffness. The actuator has a low mechanical impedance compared to geared motors, which enables it to achieve efficient interaction. Control strategies of an assistive system with the AP2A are discussed in the second part. This control framework utilizes the characteristics of the AP2A to provide assistance when necessary and to operate transparently (i.e., neither to assist nor to disturb the users) otherwise. Energy consumed by the AP2A and the assisted system is minimized by solving an optimal control problem. Finally, an estimator is introduced to detect assistive timing for the assistive system with the AP2A. This estimator utilizes physiological signals such as surface electromyogram and prior knowledge of a muscular model, classifying if the user is under the specified condition to be assisted by the AP2A. It demonstrates that the user's effort can be saved, also reducing the number of procedures to collect training data for the estimator before using assistive systems. The performance of the actuator, the controller, and the estimator proposed in this dissertation are verified through experiments.

From the above, this dissertation contributes to developing the AP2A that provides assistance and saves energy usage of assistive systems by working as a mechanical spring with stiffness optimized for achieving effective interaction under specific conditions. This actuator supports assistive devices that can be deployed in the real world, properly assisting the users when needed.