| dc.description.abstract | In the past decade, powered robotic lower-limb exoskeletons have emerged as an exciting technological advancement, offering enhancements in mobility and quality of life for individuals with physical impairments. While the potential benefits of these systems are vast, the practical implementation of exoskeletons is not without its challenges: it is difficult to remain upright and balanced without assistance from a device like crutches. Therefore, the current dissertation investigated how to address this concern through computational methods and experimental testing. To tackle the issue of poor exoskeleton balance and to address the relatively limited existing work on integrated human-biomechantronic models, especially those specific to lower-limb exoskeletons, an approach centered on predictive multibody dynamic computer simulations, full-state feedback control, optimal control, and optimization was undertaken. The research goals were broadly organized into four categories: i) Generation of novel simulations of human-only feet-in-place balance recovery in the sagittal plane; ii) Creation of an integrated symbolic model of human-exoskeleton multibody dynamics that does not rely on rigid coupling between human and exoskeleton; iii) Generation of novel dynamic simulations of human-exoskeleton balance with and without the presence of novel feet-in-place balance assistance; iv) Testing the proposed assistance algorithms in experiments with healthy young adults. The human model was represented in the sagittal plane as a 5 degree-of-freedom inverted pendulum driven by computationally-efficient and C2 continuous muscle torque generators. The exoskeleton model was designed as a floating-base biped with 9 degrees of freedom; dynamic parameters were identified separately using a combination of physical experiments and optimization approaches. Coupling the human and exoskeleton models required the development of a constitutive model of compliant human-exoskeleton interactions, in addition to a rigid-coupling formulation for efficient control-oriented model development. Predictive open-loop and closed-loop simulations of human motion were produced using a symbolic direct-collocation approach; the closed loop method introduced a novel model of central nervous system behaviour via nonlinear model predictive control. Open-loop simulations of human-only and human-exoskeleton feet-in-place balance were used to assess how human physiology and device hardware could impact the end-user's capacity for responding to external perturbations without requiring a step or falling (the feasible stability region). Novel methods for feet-in-place balance recovery assistance were derived using optimal full-state-feedback with constraints (model predictive control) paired with a nonlinear moving horizon estimator of joint angles, velocities, motor torques, and external torques ― all from noisy onboard measurements. Findings from simulations were important in providing a mechanical assessment of human and human-exoskeleton feet-in-place balance recovery, i.e., how certain mechanical qualities impact system motion and balance regulation. For example, optimized arm motion did not provide an improvement in the size of human-only feet-in-place feasible stability region. Additionally, minimizing “effort” during estimations of the stability region suggested that an ankle strategy is more efficient than a hip strategy, differing from existing literature. Furthermore, the simulation of human-only motion using nonlinear model predictive control suggested that performance criteria featuring displacements of the center of mass (COM) and trunk pitch (from reference values) prompted different human postural strategies (this was consistent with open-loop simulations). Wearing an exoskeleton, while advantageous in altering net mass distribution, did reduce the size of the end-user stability region indicating a lower capacity for responding to perturbations when wearing the device. Parameter identification using physical experiments and inverse dynamics-based optimization approaches both provided reasonable replication of experimental joint angles and actuator torques, though a final validation was difficult; the accuracy of the onboard sensor measurements must be assessed. Dynamics simulations also provided a means of assessing the capabilities of different balance regulating control algorithms. Simulations employing various full-state feedback control laws demonstrated their feasibility in introducing assistive or autonomous balancing torques for the hip, knee, and ankle. Notably, the model predictive approach outperformed others by providing optimal regulation while considering important constraints, e.g., regulating the zero-moment point in the support polygon. Lastly, experimental results indicated that friction compensation alone was sufficient for reasonable feet-in-place balance control when the end-user was a healthy young adult; the proposed COM assist-as-needed algorithm proved less effective. | en |
