Time-Invariant, Multi-Objective Extremum Seeking Control for Model-Free Auto-Tuning of Powered Prosthetic Legs

点击翻译按钮获取中文摘要

This research project seeks fundamental knowledge and understanding of versatile, adaptive optimization methods to enable real-time auto-tuning of powered prosthetic legs. Even with the help of modern prosthetic legs, lower-limb amputees often experience reduced mobility, leading to reduced quality of life and additional health problems. Recently developed powered prosthetic legs have the potential to improve outcomes, but these devices have not been clinically adopted because of the technical expertise and excessive time and effort required to configure their control systems for each patient. These control systems involve dozens of non-intuitive parameters that are specific to each user's physiology, how they walk, and environmental conditions, which also prevents these devices from adapting to the changing rhythms of daily life. Powered prostheses that automatically adjust to changing user activity and environmental conditions could significantly improve mobility for over a million lower-limb amputees in the United States alone. Furthermore, self-tuning could help powered prosthetic legs to adapt to natural changes in the patient perhaps, for example, due to fatigue. The self-tuning algorithms would have applications in control of other repetitive processes, such as powered orthoses for stroke patients, energy-harvesting turbines, HVAC systems, and biological processes. To promote knowledge transfer, the PIs will sponsor senior design projects for undergraduate student teams to design and build new experimental test beds for the developed control systems.The major objective of this research concerns novel methods of model-free adaptive optimization for systems with varying time-scales and competing objectives. Extremum seeking control (ESC) is a powerful approach to model-free adaptive optimization that requires the plant and ESC dynamics to have separated, fixed time-scales in order to optimize a single objective function. However, human locomotion exhibits varying time-scales based on activity (e.g., walking speed) and involves optimization of multiple competing objectives (e.g., energetic efficiency vs. stability). A time-invariant, multi-objective ESC framework is therefore needed to auto-tune powered prosthetic legs, which currently require several hours of customization by an expert, just for baseline operation. The overall goals of this project are to first to understand how to perform ESC of rhythmic processes with varying time-scales for real-time, model-free adaptation, next to understand how to automatically optimize multiple competing objectives using ESC, and, finally, to understand how to auto-tune a powered prosthetic leg for patient-specific behavior without a model of the human user.