Yasuko Namikawa ¹ Hiroaki Kawamoto ^2,3* Akira Uehara ^2,3 Yoshiyuki Sankai ^2,3,4

• ¹ Graduate School of Science and Technology, University of Tsukuba, Tsukuba, Ibaraki, Japan
• ² Institute of Systems and Information Engineering, University of Tsukuba, Tsukuba, Ibaraki, Japan
• ³ Center for Cybernics Research, University of Tsukuba, Tsukuba, Ibaraki, Japan
• ⁴ Cyberdyne Inc., Tsukuba, Ibaraki, Japan

The wearable cyborg Hybrid Assistive Limb (HAL) is a therapeutic exoskeletal device that provides voluntary gait assistance using kinematic/kinetic gait data and bioelectrical signals. By accumulating the gait data automatically measured by the HAL, we are developing a system to analyze gait during the intervention, which facilitates the evaluation and monitoring of treatment. This system enables analysis of the wearer's gait during HAL intervention, unlike conventional evaluations that compare gait test results before and after treatment, with minimal burden and high frequency. Despite the potential use of the gait data from the HAL's sensor information during the intervention, there is still a lack of analysis using such gait data and knowledge of gait patterns while using the HAL. Therefore, we clustered gait patterns into subgroups based on the gait data that the HAL automatically collected during treatment to investigate their characteristics. The gait data analyzed consisted of kinematic and kinetic data including ground reaction forces, joint angles, trunk angles, and HAL joint torques in individuals with progressive neuromuscular diseases. For each measured item, principal component analysis was applied to the gait time-series data to extract the features of the gait patterns, followed by hierarchical cluster analysis to generate subgroups based on the principal component scores. The gait patterns of 13710 gait cycles from 457 treatments among 48 individuals were divided into 5-10 clusters for each measured item. The clusters revealed a variety of gait patterns when wearing the HAL and identified the characteristics of multiple sub-group types. Additionally, the Bayesian regression model explained the influence of the wearer's disease type and gait ability on the distribution of gait patterns to subgroups. Our results demonstrate the importance of monitoring HAL-assisted walking to provide appropriate interventions for different gait conditions. Furthermore, our approach highlights the usefulness of the gait data that HAL automatically measures during the intervention. We anticipate that the HAL, developed as a therapeutic device, will expand its role as data measurement device for analysis and evaluation that provides gait data simultaneously with interventions, creating a novel system of cybernics treatment allowing for multi-faceted gait understanding.