6 7 I. KEYWORDS 8 Ankle exoskeleton; data-driven modeling; locomotion; prediction; joint kinematics; muscle activity 9 10 II. ABSTRACT 11Despite recent innovations in exoskeleton design and control, predicting subject-specific impacts of 12 exoskeletons on gait remains challenging. We evaluated the ability of three subject-specific phase-varying 13 models to predict kinematic and myoelectric response to ankle exoskeletons during walking, without 14 knowledge of specific user characteristics. Each modela purely phase-varying (PV), linear phase-varying 15 (LPV), and nonlinear phase-varying (NPV) modelleveraged Floquet Theory to predict deviations from a 16 nominal gait cycle due to exoskeleton torque, though the models differed in structure and expected prediction 17 accuracy. For twelve unimpaired adults walking with bilateral passive ankle exoskeletons, we predicted 18 kinematics and muscle activity in response to exoskeleton torque. Prediction accuracy of the LPV model was 19 better than the PV model when predicting less than 12.5% of a stride in the future and explained 55-70% of 20 the variance in hip, knee, and ankle kinematic responses to torque. The LPV model also predicted kinematic 21 responses with similar accuracy to the more-complex NPV model. Myoelectric responses were challenging 22 to predict across models, explaining at most 10% of the variance in responses. This work highlights the 23 potential of linear phase-varying models to predict complex responses to ankle exoskeletons and inform 24 device design. 25 42 myoelectric responses to the exoskeletons. More physiologically-detailed musculoskeletal models have been 43 used to predict the impacts of exoskeleton design on muscle activity during walking in children with cerebral 44 palsy and running in unimpaired adults [14,15]. While these studies identified hypothetical relationships 45 between kinematics and the myoelectric impacts of exoskeleton design parameters, sensitivity to 46 musculotendon properties likely limited the specificity of their predictions [16,17]. 47 48Challenges to accurately predicting responses to ankle exoskeletons with physics-based models largely stem 49 from uncertainty in adaptation, musculoskeletal physiology, and motor control, which may vary between 50 3 individuals and influence response to exoskeletons. One study found that, while individuals explore different 51 gait patterns to identify an energetically-optimal gait, exploration does not always occur spontaneously, 52 resulting in sub-optimal gait patterns for some users [18]. Popular physiologically-detailed models of human 53 gait typically assume instantaneous and optimal adaptation, which do not reflect how experience and 54 exploration may influence an individual's response to exoskeletons, possibly reducing the accuracy of 55 predicted responses [19,20]. Additionally, when specific measurement sets are unavailable for model 56 parameter tuning, models use population-average based assumptions about musculoskeletal properties and 57 motor control [17,18,[20][21...