Measurement and Validation of User-Perceived Comfort of an Ankle Exoskeleton

Restricted (Penn State Only)
- Author:
- Mohammed El Husaini, Mohammed
- Graduate Program:
- Mechanical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- October 24, 2022
- Committee Members:
- Anne Elizabeth Martin, Thesis Advisor/Co-Advisor
Bo Cheng, Committee Member
Mary Frecker, Program Head/Chair
Katie Fitzsimons, Committee Member - Keywords:
- Pneumatic
Exoskeleton
Ankle
1 DOF
Actuated
Comfort
VAS
Visual Analogue Scale
Likert
Survey
Perception
Impedance Control
Stiffness
Gait
Walking
Repeatability
Modeling
Subjective Feelings
Validation
Measurement
Rehabilitation - Abstract:
- For gait rehabilitation and industrial exoskeletons, comfort is one of the main factors for user acceptance. However, current gait-related exoskeleton research focuses on other aspects, such as minimizing the metabolic cost of transport. Unlike comfort, the methods for measuring and estimating metabolic cost are reliable and well-established for gait analysis. Therefore, comfort cannot receive the same scrutiny in exoskeleton research without first developing a validated method to measure it. A key difference between comfort and metabolic cost is that comfort is a subjective feeling which is difficult to measure partially due to limitations of speech and variances in personal interpretations among individuals. One method to measure comfort could be the use of a Visual Analogue Scale (VAS). The VAS is a type of survey that can reliably measure general pain, which is a subjective feeling. This thesis aimed to validate a VAS that measures user-perceived comfort as an ankle exoskeleton’s control parameters vary. 14 healthy subjects (12 males, 2 females, mean ± SD, age 22.6 ± 3.0 years, mass 70.1 ± 12.1 kg, and height 172 ± 11.3 cm) walked slowly on a treadmill for 3-minute trials wearing a pair of ankle exoskeletons as their biological signals were recorded. Each trial consisted of a different set of exoskeleton control parameters that changed as subjects completed at least 20, 3-minute trials during the 3-hour experiment. Each set of exoskeleton control parameters was assessed twice. After each trial, user-perceived comfort was measured using a two-question, modified 100mm VAS. However, validation of the VAS focused on the first question’s (Q1) responses: “How comfortable are you?”. The previous trial’s response was provided to subjects during measurement. The validity of the VAS was assessed by computing the repeatability coefficient (RC). In addition, to evaluate how consistently subjects rated each condition, the comfort responses were relabeled as “higher”, “lower”, or “no change” depending on the previous trial’s response. Then, the relabeled responses were compared as conditions were repeated. Furthermore, the VAS was converted into a Likert scale to assess whether the interpretation of the RC validation results changed. Lastly, QR models were constructed to validate the VAS based on the assumption that moderate control parameters would be the most comfortable. The confidence intervals (CI) of the quadratic terms were computed to determine whether they were strictly negative. The validation methods previously discussed for the VAS yielded poor results. The RC of the combined data was 39mm, and the median RC for all subjects was 33mm for Q1. Most subjects’ individual RC encompassed the majority, and in some cases, the entirety of their range of responses for each question. The results of the pairwise comparisons analysis agreed with the large RC since only 48% of the trials had consistent, repeated comfort responses. A similar conclusion was reached based on the Likert scale conversion. Furthermore, none of the quadratic models had statistically significant negative quadratic terms. The results indicated that the VAS could not repeatedly measure comfort. The large value of the RC and the inconsistent pairwise comparisons suggested that comfort perception may change as the experiment progresses. If so, the methods used to validate the VAS were inappropriate since the data violated key assumptions. More complicated models may work better, such as those including biological signals. Another possibility is that subjects could not distinguish and/or recall how comfortable previous trials were and had difficulty accurately assigning an absolute comfort score.