Stroke occurs when the blood supply to part of the brain is interrupted or reduced, preventing brain tissue from getting oxygen and nutrients, thus causing brain cells to die. Stroke is the 5th leading cause of death in the United States and is one of the major causes of disability. Conventional therapy is a form of stroke rehabilitation generally consisting of physical and occupational therapy. It focuses on customized exercises based on the patient’s feedback. Physical therapy includes exercises such as weight bearing (affected arm), vibration of affected muscle and gravity-eliminated movement of affected arm. Overall physical therapy aims at strengthening muscle groups and aides in the relearning process. Occupational aspect of conventional therapy includes activities of daily living (ADL) such as dressing, self-feeding, grooming and toileting. Overall occupational therapy focuses on improving the daily activities performed by individuals. In comparison to conventional therapy, robotic therapy is relatively newer therapy. It uses robotic devices to perform repetitive motions and delivers high dosage and high intensity training to stroke patients. Based on the research studies reviewed, it is known that neuroplasticity in stroke patients is linked to interventions which are high in dosage, intensity, repetition, difficulty, salience. Peer-reviewed literature suggests robotic therapy might be a viable option for recovery in stroke patients. However, the extent to which robotic therapy may provide greater benefits than conventional therapy remains unclear. This thesis addresses the key components of a study design for comparing the efficacy of robotic therapy relative to conventional therapy to improve upper limb sensorimotor function in stroke survivors. The study design is based on an extensive review of the literature of stroke clinical trials and robotic therapy studies, analyses of the capabilities of a robotic therapy device (M2, Fourier Intelligence), and pilot data collected on healthy controls to create a pipeline of tasks and analyses to extract biomarkers of sensorimotor functional changes. This work has laid the foundation for a pilot longitudinal study that will be conducted at the Barrow Neurological Institute, Phoenix, AZ, where conventional and robotic therapy will be compared in a small cohort of stroke survivors.

Motor learning is the process of improving task execution according to some measure of performance. This can be divided into skill learning, a model-free process, and adaptation, a model-based process. Prior studies have indicated that adaptation results from two complementary learning systems with parallel organization. This report attempted to answer the question of whether a similar interaction leads to savings, a model-free process that is described as faster relearning when experiencing something familiar. This was tested in a two-week reaching task conducted on a robotic arm capable of perturbing movements. The task was designed so that the two sessions differed in their history of errors. By measuring the change in the learning rate, the savings was determined at various points. The results showed that the history of errors successfully modulated savings. Thus, this supports the notion that the two complementary systems interact to develop savings. Additionally, this report was part of a larger study that will explore the organizational structure of the complementary systems as well as the neural basis of this motor learning.

Transcranial focused ultrasound (tFUS) is a unique neurostimulation modality with potential to develop into a highly sophisticated and effective tool. Unlike any other noninvasive neurostimulation technique, tFUS has a high spatial resolution (on the order of millimeters) and can penetrate across the skull, deep into the brain. Sub-thermal tFUS has been shown to induce changes in EEG and fMRI, as well as perception and mood. This study investigates the possibility of using tFUS to modulate brain networks involved in attention and cognitive control.Three different brain areas linked to saliency, cognitive control, and emotion within the cingulo-opercular network were stimulated with tFUS while subjects performed behavioral paradigms. The first study targeted the dorsal anterior cingulate cortex (dACC), which is associated with performance on cognitive attention tasks, conflict, error, and, emotion. Subjects performed a variant of the Erikson Flanker task in which emotional faces (fear, neutral or scrambled) were displayed in the background as distractors. tFUS significantly reduced the reaction time (RT) delay induced by faces; there were significant differences between tFUS and Sham groups in event related potentials (ERP), event related spectral perturbation (ERSP), conflict and error processing, and heart rate variability (HRV).
The second study used the same behavioral paradigm, but targeted tFUS to the right anterior insula/frontal operculum (aIns/fO). The aIns/fO is implicated in saliency, cognitive control, interoceptive awareness, autonomic function, and emotion. tFUS was found to significantly alter ERP, ERSP, conflict and error processing, and HRV responses.
The third study targeted tFUS to the right inferior frontal gyrus (rIFG), employing the Stop Signal task to study inhibition. tFUS affected ERPs and improved stopping speed. Using network modeling, causal evidence is presented for rIFG influence on subcortical nodes in stopping.
This work provides preliminarily evidence that tFUS can be used to modulate broader network function through a single node, affecting neurophysiological processing, physiologic responses, and behavioral performance. Additionally it can be used as a tool to elucidate network function. These studies suggest tFUS has the potential to affect cognitive function as a clinical tool, and perhaps even enhance wellbeing and expand conscious awareness.

The ability to inhibit a planned but inappropriate response, and switch to executing a goal-relevant motor response, is critically important for the regulation of motor behaviors. Inhibition and switching could be mediated by various control mechanisms. Proactive control uses contextual information (cues) to plan the response for the target stimulus (probe) based on the expectation of a response inhibition or switching stimulus combination. Previous work has reported the involvement of several brain areas associated with proactive inhibition and switching, e.g., dorsolateral prefrontal cortex, anterior cingulate cortex, inferior frontal junction, and pre-supplementary motor area. However, how these areas interact and their functional role in different types of cognitive control is still debated. An AX-version of the continuous performance task (AX-CPT) was used to examine proactive inhibition and switching of motor actions. In a typical AX-CPT trial, a contextual cue stimulus is presented, followed by a probe stimulus after a specific inter-stimulus interval. As part of a trial sequence, if a target cue and target probe are presented, a target response is to be provided when the probe is observed. Otherwise, a non-target response is to be provided for all other stimuli. A behavioral switching AX-CPT experiment (48 subjects) was conducted to explore the parameters that induce a proactive shift in the motor response. Participants who performed the AX-CPT task with relatively shorter interstimulus interval predominantly and consistently exhibited proactive control behavior. A follow-up pilot study (3 subjects) of response inhibition versus response switching AX-CPT was performed using 256-channel high-density electroencephalography (HD-EEG). HD-EEG was used to identify the time course of cortical activation in brain areas associated with response inhibition. It was observed that one out of three participants used a proactive strategy for response switching based on probe response error and probe response reaction time. Instantaneous amplitude spatial maps obtained from HD-EEG revealed cortical activity corresponding to conflict between proactively-prepared incorrect responses and reactively-corrected goal-relevant responses after the probe was presented.

The term Poly-Limb stems from the rare birth defect syndrome, called Polymelia. Although Poly-Limbs in nature have often been nonfunctional, humans have had the fascination of functional Poly-Limbs. Science fiction has led us to believe that having Poly-Limbs leads to augmented manipulation abilities and higher work efficiency. To bring this to life however, requires a synergistic combination between robot manipulation and wearable robotics. Where traditional robots feature precision and speed in constrained environments, the emerging field of soft robotics feature robots that are inherently compliant, lightweight, and cost effective. These features highlight the applicability of soft robotic systems to design personal, collaborative, and wearable systems such as the Soft Poly-Limb.

This dissertation presents the design and development of three actuator classes, made from various soft materials, such as elastomers and fabrics. These materials are initially studied and characterized, leading to actuators capable of various motion capabilities, like bending, twisting, extending, and contracting. These actuators are modeled and optimized, using computational models, in order to achieve the desired articulation and payload capabilities. Using these soft actuators, modular integrated designs are created for functional tasks that require larger degrees of freedom. This work focuses on the development, modeling, and evaluation of these soft robot prototypes.

In the first steps to understand whether humans have the capability of collaborating with a wearable Soft Poly-Limb, multiple versions of the Soft Poly-Limb are developed for assisting daily living tasks. The system is evaluated not only for performance, but also for safety, customizability, and modularity. Efforts were also made to monitor the position and orientation of the Soft Poly-Limbs components through embedded soft sensors and first steps were taken in developing self-powered compo-nents to bring the system out into the world. This work has pushed the boundaries of developing high powered-to-weight soft manipulators that can interact side-by-side with a human user and builds the foundation upon which researchers can investigate whether the brain can support additional limbs and whether these systems can truly allow users to augment their manipulation capabilities to improve their daily lives.

Tactile and proprioceptive sensory feedback are the two sensory modalities that make up haptic sensation. The degree which these two sensory modalities are integrated together is not very well known. To investigate this issue a set of experiments were set into motion separating these sensory modalities and testing what happens when a person’s proprioceptive system is perturbed. A virtual reality system with haptic feedback along with a weighted object were utilized in a reach, grasp, and lift task. The subjects would lift two objects sequentially and try to judge which one was heavier. This project was split into three different experiments to measure the subject’s perception in different situations. The first experiment utilized the virtual reality system to measure the perception when the subject only has proprioceptive inputs. The second experiment would include the virtual reality system and the weighted object to act as a comparison to the first experiment with the additional tactile input. The third experiment would then add perturbations to the proprioceptive inputs through the virtual reality system to investigate how perception will change. Results from experiment 1 and 2 showed that subjects are almost just as accurate with weight discrimination even if they only have proprioceptive inputs however, subjects are much more consistent in their weight discrimination with both sensory modalities. Results from experiment 3 showed that subjective perception does change when the proprioception is perturbed but the magnitude of that change in perception depends on the perturbation performed.

The human ankle is a vital joint in the lower limb of the human body. As the point of interaction between the human neuromuscular system and the physical world, the ankle plays important role in lower extremity functions including postural balance and locomotion . Accurate characterization of ankle mechanics in lower extremity function is essential not just to advance the design and control of robots physically interacting with the human lower extremities but also in rehabilitation of humans suffering from neurodegenerative disorders.

In order to characterize the ankle mechanics and understand the underlying mechanisms that influence the neuromuscular properties of the ankle, a novel multi-axial robotic platform was developed. The robotic platform is capable of simulating various haptic environments and transiently perturbing the ankle to analyze the neuromechanics of the ankle, specifically the ankle impedance. Humans modulate ankle impedance to perform various tasks of the lower limb. The robotic platform is used to analyze the modulation of ankle impedance during postural balance and locomotion on various haptic environments. Further, various factors that influence modulation of ankle impedance were identified. Using the factors identified during environment dependent impedance modulation studies, the quantitative relationship between these factors, namely the muscle activation of major ankle muscles, the weight loading on ankle and the torque generation at the ankle was analyzed during postural balance and locomotion. A universal neuromuscular model of the ankle that quantitatively relates ankle stiffness, the major component of ankle impedance, to these factors was developed.

This neuromuscular model is then used as a basis to study the alterations caused in ankle behavior due to neurodegenerative disorders such as Multiple Sclerosis and Stroke. Pilot studies to validate the analysis of altered ankle behavior and demonstrate the effectiveness of robotic rehabilitation protocols in addressing the altered ankle behavior were performed. The pilot studies demonstrate that the altered ankle mechanics can be quantified in the affected populations and correlate with the observed adverse effects of the disability. Further, robotic rehabilitation protocols improve ankle control in affected populations as seen through functional improvements in postural balance and locomotion, validating the neuromuscular approach for rehabilitation.

Due to the advent of easy-to-use, portable, and cost-effective brain signal sensing devices, pervasive Brain-Machine Interface (BMI) applications using Electroencephalogram (EEG) are growing rapidly. The main objectives of these applications are: 1) pervasive collection of brain data from multiple users, 2) processing the collected data to recognize the corresponding mental states, and 3) providing real-time feedback to the end users, activating an actuator, or information harvesting by enterprises for further services. Developing BMI applications faces several challenges, such as cumbersome setup procedure, low signal-to-noise ratio, insufficient signal samples for analysis, and long processing times. Internet-of-Things (IoT) technologies provide the opportunity to solve these challenges through large scale data collection, fast data transmission, and computational offloading.

This research proposes an IoT-based framework, called BraiNet, that provides a standard design methodology for fulfilling the pervasive BMI applications requirements including: accuracy, timeliness, energy-efficiency, security, and dependability. BraiNet applies Machine Learning (ML) based solutions (e.g. classifiers and predictive models) to: 1) improve the accuracy of mental state detection on-the-go, 2) provide real-time feedback to the users, and 3) save power on mobile platforms. However, BraiNet inherits security breaches of IoT, due to applying off-the-shelf soft/hardware, high accessibility, and massive network size. ML algorithms, as the core technology for mental state recognition, are among the main targets for cyber attackers. Novel ML security solutions are proposed and added to BraiNet, which provide analytical methodologies for tuning the ML hyper-parameters to be secure against attacks.

To implement these solutions, two main optimization problems are solved: 1) maximizing accuracy, while minimizing delays and power consumption, and 2) maximizing the ML security, while keeping its accuracy high. Deep learning algorithms, delay and power models are developed to solve the former problem, while gradient-free optimization techniques, such as Bayesian optimization are applied for the latter. To test the framework, several BMI applications are implemented, such as EEG-based drivers fatigue detector (SafeDrive), EEG-based identification and authentication system (E-BIAS), and interactive movies that adapt to viewers mental states (nMovie). The results from the experiments on the implemented applications show the successful design of pervasive BMI applications based on the BraiNet framework.

It has been repeatedly shown that females have lower stability and increased risk of ankle injury when compared to males participating in similar sports activities (e.g., basketball and soccer), yet sex differences in neuromuscular control of the ankle, including the modulation of ankle stiffness, and their contribution to stability remain unknown. To identify sex differences in human ankle stiffness, this study quantified 2- dimensional (2D) ankle stiffness in 20 young, healthy men and 20 young, healthy women during upright standing over a range of tasks, specifically, ankle muscle co-contraction tasks (4 levels up to 20% maximum voluntary co-contraction of ankle muscles), weight-bearing tasks (4 levels up to 90% of body weight), and ankle torque generation tasks accomplished by maintaining offset center-of-pressure (5 levels up to +6 cm to the center-of-pressure during quiet standing). A dual-axial robotic platform, capable of perturbing the ankle in both the sagittal and frontal planes and measuring the corresponding ankle torques, was used to reliably quantify the 2D ankle stiffness during upright standing. In all task conditions and in both planes of ankle motion, ankle stiffness in males was consistently greater than that in females. Among all 26 experimental conditions, all but 2 conditions in the frontal plane showed statistically significant sex differences. Further analysis on the normalized ankle stiffness scaled by weight times height suggests that while sex differences in ankle stiffness in the sagittal plane could be explained by sex differences in anthropometric factors as well as neuromuscular factors, the differences in the frontal plane could be mostly explained by anthropometric factors. This study also demonstrates that the sex differences in the sagittal plane were significantly higher as compared to those in the frontal plane. The results indicate that females have lower ankle stiffness during upright standing thereby providing the neuromuscular basis for further investigations on the correlation of ankle stiffness and the higher risk of ankle injury in females.

Robotic lower limb prostheses provide new opportunities to help transfemoral amputees regain mobility. However, their application is impeded by that the impedance control parameters need to be tuned and optimized manually by prosthetists for each individual user in different task environments. Reinforcement learning (RL) is capable of automatically learning from interacting with the environment. It becomes a natural candidate to replace human prosthetists to customize the control parameters. However, neither traditional RL approaches nor the popular deep RL approaches are readily suitable for learning with limited number of samples and samples with large variations. This dissertation aims to explore new RL based adaptive solutions that are data-efficient for controlling robotic prostheses.

This dissertation begins by proposing a new flexible policy iteration (FPI) framework. To improve sample efficiency, FPI can utilize either on-policy or off-policy learning strategy, can learn from either online or offline data, and can even adopt exiting knowledge of an external critic. Approximate convergence to Bellman optimal solutions are guaranteed under mild conditions. Simulation studies validated that FPI was data efficient compared to several established RL methods. Furthermore, a simplified version of FPI was implemented to learn from offline data, and then the learned policy was successfully tested for tuning the control parameters online on a human subject.

Next, the dissertation discusses RL control with information transfer (RL-IT), or knowledge-guided RL (KG-RL), which is motivated to benefit from transferring knowledge acquired from one subject to another. To explore its feasibility, knowledge was extracted from data measurements of able-bodied (AB) subjects, and transferred to guide Q-learning control for an amputee in OpenSim simulations. This result again demonstrated that data and time efficiency were improved using previous knowledge.

While the present study is new and promising, there are still many open questions to be addressed in future research. To account for human adaption, the learning control objective function may be designed to incorporate human-prosthesis performance feedback such as symmetry, user comfort level and satisfaction, and user energy consumption. To make the RL based control parameter tuning practical in real life, it should be further developed and tested in different use environments, such as from level ground walking to stair ascending or descending, and from walking to running.