Towards Efficient Personalization of Computerized Lower Limb Prostheses via Reinforcement Learning in a Clinical Setup - PROJECT SUMMARY Although lower limb (LL) prosthetics can restore basic mobility, not being able to handle different locomotion tasks has been the top complaint from patients about the traditional passive prosthetic legs. Recently, advanced powered prosthetic legs have become clinically available. They have been shown assisting various locomotion tasks, improving walking efficiency, lessening undue compensation from intact joints, and reducing secondary injuries (such as back pain). However, none of those benefits can be guaranteed until the control of powered LL prostheses is personalized appropriately. Current clinical practice in personalizing powered LL prostheses has been performed manually and heuristically. This approach is time and labor intensive. Due to time restrictions in clinical visits, prosthetists could only personalize a subset of prosthesis control parameters, limiting the device tuning precision and the user’s locomotion functions. In addition, manufacturers need to train prosthetists with specialized knowledge about control of a specific prosthesis prior to personalizing the device for a patient. Since the number of these specialists is limited, accessing their clinics is challenging and costly to LL amputees. Therefore, a new solution to transform the current clinical practice is urgently needed. To address this clinical need, our team has pioneered reinforcement learning (RL)-based approach that can simultaneously personalize high-dimension prosthesis control parameters automatically, quickly, and safely. Yet, how to translate this innovative method into clinics, and whether it benefits the prosthetists and amputee users have not been investigated. The objective is to investigate the efficacy of our RL method in prosthesis personalization clinics. We propose a RL-based Recommendation Interfacing System (RISE) that automatically recommends the prosthesis control parameters to prosthetists to meet their device tuning needs, which are based on their individual clinical judgment and the user’s verbal feedback. Our central hypothesis is that compared to current clinical practice, RISE will significantly improve the work efficiency of a prosthetist, i.e., they can perform personalization of any powered prosthesis quickly and accurately without a need to learn engineering control of powered prostheses. This project is clinically significant, because not only it improves the efficiency of clinical practice, but also it increases LL amputee users’ accessibility to powered prosthesis and its tuning clinics, improves their locomotion functions, and reduces the cost to amputees associated with the powered prosthesis use.
