Informing Ankle-Foot Prosthesis Prescription through Haptic Emulation of Candidate Devices Joshua M. Caputo 1,2,* , Peter G. Adamczyk 2 , Steven H. Collins 1,3 Abstract— Robotic prostheses can improve walking perfor- mance for amputees, but prescription of these devices has been hindered by their high cost and uncertainty about the degree to which individuals will benefit. The typical prescription process cannot well predict how an individual will respond to a device they have never used because it bases decisions on subjective assessment of an individual’s current activity level. We propose a new approach in which individuals ‘test drive’ candidate devices using a prosthesis emulator while their walking performance is quantitatively assessed and results are distilled to inform prescription. In this system, prosthesis behavior is controlled by software rather than mechanical implementation, so users can quickly experience a broad range of devices. To test the viability of the approach, we developed a prototype emulator and assessment protocol, leveraging hard- ware and methods we previously developed for basic science experiments. We demonstrated emulations across the spectrum of commercially available prostheses, including traditional (e.g. SACH), dynamic-elastic (e.g. FlexFoot), and powered robotic (e.g. BiOM R  T2) prostheses. Emulations exhibited low er- ror with respect to reference data and provided subjectively convincing representations of each device. We demonstrated an assessment protocol that differentiated device classes for each individual based on quantitative performance metrics, providing feedback that could be used to make objective, personalized device prescriptions. I. I NTRODUCTION A. Typical Prescription Process The prescription of ankle-foot prostheses is hindered by uncertainty about which device is most suitable for a given individual [1]. Payers expect justification for prosthesis se- lection, but without objective data clinicians can only pro- vide their subjective impression, the expressed needs of the individual, and, at best, basic assessment of an individual’s pre-prescription mobility [2]. Recent robotic devices have intensified this problem, as they have demonstrated benefits to the user [3, 4], but at a high price (about $80,000 for a BiOM R  T2 vs. about $1,000 for a conventional prosthesis). The degree to which individual users will benefit also re- mains unclear. Given this uncertainty, clinical practice is slow to accommodate disruptive technologies, and is not able to effectively predict a user’s activity-level and ability with a device they have never used. This material is based upon work supported by the National Institutes of Health under Award No. 1R43HD076518-01 and is the subject of US Provisional Patent No. 62/070,134. The authors are affiliated with Intelligent Prosthetic Systems, LLC, which is pursuing commercialization of the prosthesis emulator system. 1 Department of Mechanical Engineering, Carnegie Mellon University 2 Intelligent Prosthetic Systems, LLC 3 Robotics Institute, Carnegie Mellon University ∗ Corresponding author, joshua.m.caputo@gmail.com B. Informing Prescription by Haptic Emulation We propose a new approach, wherein patients ‘test drive’ candidate devices, providing hard data on how they perform with each prosthesis. This could be done by buying and trying many different prostheses for each individual, but the process would be laborious and would require expensive inventories of different models of prosthesis (each with variations for different body weights, activity levels, and foot sizes). Instead, clinicians could fit patients with a prosthesis emulator and provide the experience of wearing these dif- ferent prostheses by simply switching modes in a software interface. Most commercially-available devices can be clas- sified into one of three groups: traditional stiff and dissaptive solid ankle cushioned heel (SACH) prostheses, conventional spring-like dynamic elastic response (DER) prostheses, and actively-controlled robotic prostheses. Emulating these di- verse behaviors with a single prosthesis requires versatility beyond the capabilities of currently-available mobile robotic prostheses, which are fine-tuned to exhibit specific behaviors in a convenient autonomous package. To maximize versatility in basic science experiments that do not require autonomy, e.g. [5], we previously developed a robotic prosthesis system in which a powerful off-board motor and controller actu- ate a lightweight prosthesis end-effector through a flexible Bowden cable transmission [6]. In the present study we test whether such a system can convincingly emulate the behavior of existing off-the-shelf prostheses. C. Metrics for Evaluating Benefit To evaluate the benefits each emulation mode provides to an individual, it would be useful to have outcome metrics that capture aspects of performance that are relevant to daily life. The most-cited measure for the efficacy of an assistive device is metabolic rate (the rate at which biochemical energy is used by the body to perform a task). However, in clinical practice, the expensive equipment required to mea- sure metabolic rate is typically not available. Also, energy consumption must be balanced against other factors such as comfort, stability, versatility, and maximal performance. Therefore, it would be useful to have a set of outcomes that can be measured simply and quickly in a clinical setting, and can estimate energy consumption as well as other important outcomes. Heart rate scales roughly with metabolic rate [7] and could be used as a surrogate that is simpler to measure and responds more quickly to the task. Maximum sustainable walking speed (MSWS) also scales with metabolic rate [8], and might include information about perceived stability and comfort. Finally, patient-reported satisfaction scores and 2015 IEEE International Conference on Robotics and Automation (ICRA) Washington State Convention Center Seattle, Washington, May 26-30, 2015 978-1-4799-6923-4/15/$31.00 ©2015 IEEE 6445