2026 Theses Doctoral

Designing, Analyzing, and Assisting In-Place Motor Task Performance: Leveraging Virtual Reality and Robotic Assistance to Promote Motor Learning and Recovery

Carrera, Robert M.

Technologies including robotics, virtual reality, and neuromodulation have the potential to en- hance the effects of rehabilitation on motor control and function in impaired populations. Reha- bilitative robots have proven effective in improving outcomes in certain applications, such as for gait and upper extremity training post-stroke, but have yet to find widespread clinical adoption in many applications for which they show promise in the lab. Virtual reality has also shown im- proved outcomes in certain applications, but the wide variety of technologies, interventions, and measured outcomes have hindered drawing clear conclusions about how to effectively create VR- based interventions. Neuromodulation technologies, including epidural and transcutaneous spinal cord stimulation (ESCS/TSCS) and transcranial magnetic stimulation (TMS) are relatively new, although they show great promise in improving motor recovery with training and, in the case of ESCS, as assistive devices that partially restore or enhance volitional muscle activation in people with spinal cord injury.

Some common themes around how to better promote neuronal plasticity, motor learning, retention, transfer, and ultimately motor recovery and function have emerged from neuroscience and motor control research. In addition to being intensive and task-oriented, therapeutic motor tasks should be well-suited to the person’s capabilities, appropriately challenging and adaptive, variable, motivating, and should encourage an external focus of attention. Similarly, physical assistance during movement should promote volitional muscle activation, allow for movement variability, and adjust with the user over time. Robotics and VR can also provide quantifiable, well-defined motor tasks to track motor recovery over time or even to compare motor control and recovery between groups or populations. These ideas are changing robotic and VR-based rehabilitative interventions and can hopefully drive continued improvements in the effects of rehabilitation on motor recovery and function. This dissertation makes contributions towards more fully realizing the potential of VR- and robot-based therapies by engaging with and incorporating some of these concepts.

The first project described in the dissertation introduces a new framework for designing and analyzing unconstrained, skilled motor tasks in immersive VR. The framework borrows the concept of a goal function, which had previously been used to define virtual motor tasks performed with highly constrained (e.g., linear or planar) input devices, and adapts them for use in motor tasks where the person can move their hands freely to perform the task. The framework presents an approach for adapting these tasks to the performer’s reachable workspace, movement speed, and ability, with attention given to needed modifications for impaired populations. Our approach also proposes quantitative methods for analyzing task performance and how it changes over time. The results showed several key findings: performing tasks near the edge of the postural or reaching workspace worsened performance; performance degraded in the far workspace for task-specific reasons; people improved at similar rates regardless of workspace location (or, in one task, im- proved faster near the edge of the workspace); people performing a VR striking task improved in a similar way to people practicing table tennis. These results suggested our framework captured im- portant features of motor learning and could be used to describe how performance improvements were achieved with practice. The framework provides a basis for training and researching motor control in these rarely explored, skilled, unconstrained tasks.

We continued by testing several of the proposed methods for adjusting a motor task to the performer’s ability, with the goal of showing their feasibility and effect in an impaired population. Visual error manipulation (VEM) was used to adjust the perceived difficulty of a VR striking task. A control intervention and two types of VEM were applied to three groups of unimpaired subjects: a type that only adjusted visual error related to forward and lateral strike velocity or forward and vertical strike velocity, and a type that adjusted both simultaneously. Additionally, six children with cerebral palsy (CP) of various severities practiced the adjusted motor task. Preliminary results showed that, with adjustments, children with CP and Gross Motor Functional Classification System levels I-III could successfully perform the task and improve with practice.

The aim of the second project was to improve the ability of the stand trainer, a cable-driven parallel robot (CDPR) for postural training, to provide assistance for in-place postural training tasks. As a part of this effort, we hoped to improve the device’s usability and robustness, introduce more effective control strategies that could better assist the user, and integrate motor tasks specified in VR. Modifications to the robot’s software, sensors, and hardware allowed for reduced set-up times, more robustness in tracking, and for VR integration.

We then conducted a first experiment to characterize how unimpaired subjects coordinated their posture while performing maximal reaches in different directions and at different heights.The work employed analysis techniques from robotics to quantify how different joints (e.g., the ankle or lumbothoracic axial rotation) contributed to the reaching workspace. The analysis showed that the ankle only made significant direct contributions to higher lateral reaches, and otherwise played a critical, indirect role in counterbalancing the trunk mass to maximize the achieved reaching distance. The analysis also revealed the surprisingly minimal role other degrees of freedom played in maximal reaching, including lateral bending and lumbothoracic axial rotation.

With a better understanding of the requirements of postural coordination for in-place tasks, we undertook the development of a novel postural assistance approach based on a biomechanical model and gravity compensation. The model would allow for the stable application of large assis- tive forces by coordinating forces applied to different body segments. We validated the concept in an unimpaired population performing squats - coordinated pelvis and shank forces were effective at providing targeted knee extensor assistance throughout the range of motion of the squat, reduc- ing vastus medialis activation while preserving ankle muscle activation and squat kinematics. On the other hand, a pelvic force alone either had a limited assistance range (e.g., a vertical pelvic force) or substantially disturbed the squat performance. The work showed the feasibility of using a simple postural model to coordinate multisegment forces applied by a CDPR and showed the need for such coordination when applying multisegment forces. The same control approach could likely work for assisting standing postural tasks such as reaching.

The last project investigated the effects of transcutaneous spinal cord stimulation (TSCS) on postural control in unimpaired adults undergoing robotic perturbations. TSCS has been shown to impede naturally-generated action potentials in type Ia and Ib large-diameter afferents associated with proprioception, via generation of antidromic action potentials in those neurons. The work found that subjects undergoing perturbations and TSCS showed direction-specific increases in the activation of muscles not usually associated with the reactive postural response in that direction. The results suggested that TSCS disturbed reactive postural control in this unimpaired group, with implications for its use in impaired groups.

The dissertation concludes by describing likely future avenues for research, with a particular focus on model-based robotic rehabilitation and assistance.