Experimental and Analytical Methods for Understanding User Gait Biomechanics in Response to Soft Ankle Exosuits

Abstract

The field of lower-limb exoskeletons and exosuits for gait augmentation and restoration has developed several devices that apply torques in parallel with the biological joint. These systems have seen promising responses in both healthy and clinical populations, but there remains much to learn about how these devices interact with a wearer and how to best use them to both support and assess recovery. This thesis combines an understanding of high-level controls for wearable robotics, neuromotor learning in rehabilitation, and gait biomechanical data analysis, towards addressing this challenge.

A common goal for post-stroke rehabilitation is to increase propulsion generation, the force used to propel the body forward, by the more impaired, or paretic, ankle joint. Consequently, many wearable systems assist paretic propulsion by providing additional torque at the ankle. However, this approach does not encourage increased use of the individual's biological mechanisms and instead risks leading to reliance on the device. This thesis starts by extending the capabilities of the soft ankle exosuit by introducing a new resistive mode for increasing paretic propulsion generation that is sustained after training. We first validate this system in unimpaired individuals and show that participants significantly increased biological ankle torque and soleus muscle activity while training with active exosuit resistance. We further show the effect of varying resistance magnitude on compensatory gait responses in these users. Then, we demonstrate the efficacy of this paradigm to elicit the latent propulsion reserve in people post-stroke. Individuals increased paretic propulsion generation by statistically and clinically significant magnitudes both during and immediately after walking with exosuit resistance. This improvement also corresponded to significant increases in paretic biological ankle power and torque without compensatory involvement of the unresisted joints or limb. Moreover, these changes were not observed in traditional treadmill-based training, supporting the role of the resistive exosuit in achieving these outcomes.

In addition to training in controlled environments, literature suggests the importance of context-specific and continued training outside of the clinic, which requires monitoring gait outside of the lab. Thus, this thesis then develops and validates a machine learning-based method to estimate propulsion using portable body-worn sensors. Prior work primarily focuses on unimpaired populations, vertical loading metrics, and single sensing modalities. Here, we develop a method applicable for both individuals post-stroke and young healthy adults for the antero-posterior component of ground reaction force by using a combination of different sensing modalities, i.e., commercial pressure insoles and inertial measurement units (IMUs). We show that combining insole and IMU data improves performance over either the insole or IMU alone, and achieves errors near or within clinically-relevant thresholds. We further show that this multi-modal architecture improves performance in small training datasets, out-of-distribution walking speeds, and new environments compared to either mode alone. Finally, we demonstrate the use of this estimation approach to track propulsion in real-world environments as well as during two active gait training interventions, functional electrical stimulation and exosuit-based resistance.

These first two contributions expand the functionality of exosuit-based interventions to increase user involvement and enable tracking user biomechanics in real-world environments. However, variability across users results in a diverse range of biomechanical responses to wearable devices, leading to a need for individual-level parameter selection and analysis methods. Hence, this thesis concludes by exploring techniques for individualized analysis and intervention in unimpaired and post-stroke populations. We first analyze user-specific biomechanical response to exosuit-based resistance training in people post-stroke. We show that individuals adapt to the intervention by gradually transitioning from modulating propulsion with the paretic hip to the paretic ankle. Then in unimpaired users, we show that individual-specific soleus muscle dynamics can inform more versatile ankle exosuit assistance. We demonstrate significant metabolic reductions across a range of walking speeds and inclines using this approach. Finally, we introduce a method to partially automate extraction of the biomechanical changes in people post-stroke while walking with exosuit-applied assistance. By combining functional data analysis methods for signal alignment with hierarchical clustering, we reveal robust and interpretable stratification of user response strategies.

Together, this thesis provides new experimentally and analytically driven insights of how wearers respond to soft exosuits, supporting eventual improved clinic- and community-based rehabilitation.