Task-Invariant Control and Pre-clinical Validation of Partial Assist Exoskeletons

Abstract

A transition from powerful, bulky, and stiff jointed exoskeletons for driving the limbs of paralyzed individuals to lightweight, highly backdrivable, partial assist exoskeletons for assisting broad populations with mild to moderate mobility impairments is well underway. However this transition cannot be successfully completed without developing and in-vivo testing controllers that are versatile over multiple activities, clinically intuitive, and easily customizable based on each individual’s unique needs. This dissertation is focused on providing solutions to this challenging set of requirements via four aims: 1) improving and later assessing the performance (and limitations) of existing “task-invariant” controllers implemented on various backdrivable exoskeletons, 2) developing a novel bilateral knee controller for broad use cases, 3) performing in-vivo validation of the novel controller in the fatigue causing lifting-lowering-carrying tasks, and 4) exploring the customizability of the novel controller for meeting unique needs in highly impaired cases of post-polio-syndrome (PPS) and multiple sclerosis (MS). Accordingly, this work firstly improved a potential energy shaping (body weight supporting) controller by blending its stance and swing torques in multi-support gait phases, by utilizing the vertical ground reaction force signal from a custom designed foot pressure sensor. Subsequently, this controller and more advanced total energy shaping controllers underwent in-vivo testing focused on assessing muscle effort reductions. However, uncovering of shortcomings in customizability and unhelpful behavior outside the normative kinematics datasets (which these “data-driven” controllers strictly relied on) made them unsuitable for aims 3 and 4. By using physically inspired torque basis functions that were intuitively modified and “task-sensitized” to ultimately behave in a biomimetic fashion for multiple tasks, aim 2 produced a versatile, clinically intuitive, and “task-invariant” bilateral knee controller that achieved good in-silico as well as in-vivo results in pilot testing. Aim 3 utilized this novel controller on a highly backdrivable exoskeleton to achieve holistic, multifaceted (performance, postural, muscular, and perceptual) benefits in lifting-lowering-carrying over multiple terrain in both non-fatigued and highly-fatigued physical states. Finally aim 4 produced a clinician-friendly android app (GUI) that helped customize the novel controller for participants with PPS and MS. Meaningful improvements were found with the exoskeleton in the primary metrics, i.e., reductions in the 5xSTS time and stairs ascent time for the participant with PPS; and improvements in leg clearance and compensatory circumduction for the participant with MS.