Compensatory Stepping: Biomechanical Analysis, Predictive Modeling, and Assistance with Lower-Limb Exoskeleton

Abstract

Balancing is a fundamental human motor function that is continuously performed in daily life. Despite the fact that quiet standing is inherently unstable, humans are able to maintain upright posture through a combination of balance strategies in response to external disturbances. Small perturbations are typically counteracted using ankle and hip strategies, whereas larger and unexpected perturbations often require a reactive stepping response to prevent a fall. Compensatory stepping is a rapid, largely subconscious balance recovery strategy in which a step is taken when the center of mass, or its dynamic equivalent, moves beyond the base of support.

This thesis focuses on the mechanisms underlying compensatory stepping, particularly in response to anterior--posterior impulsive torso perturbations. By investigating the biomechanical and physical principles governing this behavior, the goal is to develop predictive models that can be leveraged for balance assistance in lower-limb exoskeletons, thereby supporting individuals with impaired balance control.

We first conducted human-only compensatory stepping experiments to gain biomechanical insights and identify relationships between stepping parameters (step length and step duration), perturbation characteristics, muscle activation and co-contraction patterns, ground-referenced stability measures, and critical temporal events. Exploratory and regression-based analyses revealed several trends, including relationships between perturbation impulse and stepping probability, non-heel-strike foot contacts, lack of hip balance strategy, and phase-dependent muscle co-contraction strategies. These new findings provide a deeper understanding of the coordination between neuromuscular activity and mechanical stability during reactive stepping.

Building on these insights, a physics-based stepping model was developed to derive a dynamic condition that determines when stepping is required following an impulsive perturbation. A multi-phase, multibody dynamic Compensatory Foot Placement Estimator (CFPE) was then proposed to predict step length and step duration by optimizing multiple cost functions, including stability, joint torque, and restoring effort-related terms. A genetic algorithm was employed for optimization, and the influence of individual cost terms and trajectory generation assumptions was systematically analyzed. Results indicated that the relative importance of cost functions varies with perturbation magnitude and step characteristics.

To overcome the limitations of model-based prediction from a single initial condition and errors due to model simplification, a data-driven convolutional neural network (CNN) was developed to estimate stepping parameters using pre–toe-off exoskeleton kinematic histories as inputs. The instant at which the center-of-mass projection crosses the base of support was identified as a key event containing sufficient predictive information prior to toe-off. The CNN demonstrated robust prediction performance across participants, and recursive feature elimination was used to identify the most influential kinematic inputs.

Finally, two exoskeleton-assisted compensatory stepping controllers—a scaled feedforward controller and a modified velocity flow field (VFF) feedback controller—were implemented on an Indego lower-limb exoskeleton using the predicted stepping parameters. Experimental evaluation using quantitative metrics and participant feedback highlighted the advantages and limitations of each control strategy and demonstrated the feasibility of predictive, parameter-driven balance assistance.

Overall, this work presents a comprehensive framework for understanding, modeling, and assisting compensatory stepping. The findings contribute to improved balance modeling and provide a foundation for the development of intelligent, predictive controllers for rehabilitative and assistive exoskeleton technologies.