Content area
Abstract
Multi-channel closed-loop interfaces using biosignals measured from sensors in, on, or around our bodies have the potential to revolutionize the way we interact with devices, machines, and each other. Such signals can be used to more intuitively and accessibly control human-machine interfaces (HMIs) for machines like vehicles, computing devices, or virtual reality avatars. However, we do not yet have the tools to develop individualized adaptive interfaces that are stable in closed-loop with an adaptive human. This dissertation focuses on how individualized adaptive algorithms can be leveraged to improve accessibility and health toward a more inclusive and equitable society.
A significant challenge with multi-channel biosignals-based control is that the measured signals are noisy and require the user to correct for errors in addition to tracking a desired trajectory. For example, to control a powered wheelchair with biosignals like electromyography (EMG), users must correct for noisy signals that make the wheelchair deviate from the intended path using feedback control. In parallel, users can track a desired path using feedforward control. We first need to develop methods that can separate feedforward and feedback control. As an initial step, I recruited non-disabled participants to perform trajectory-tracking and disturbance-rejection tasks and used control theory methods to separately quantify the user’s feedforward and feedback control. I used this model to examine whether handedness affects learned controllers with participants without disabilities. Participants learned feedforward and feedback controllers equally well with both hands and participants' learned controllers transferred between hands. Reduction of motor noise (i.e., extraneous movements to the task at hand) was a large factor in improved trajectory-tracking, highlighting the need for an assistive algorithm to compensate for a person’s intrinsic motor noise.
I next investigated how HMIs can be enhanced with EMG as a multi-channel control method. EMG is an attractive multi-channel, intuitive, always available control option because it is non-invasive, can be detected even if muscle contractions are small, and sensor placement can be individualized to the abilities and preferences of the user. I conducted a study with nondisabled participants comparing EMG and manual interfaces during a trajectory-tracking task when people are controlling the velocity and the acceleration of a cursor on a screen. People implemented better feedforward control (i.e., closer to the ideal controller to perfectly track the reference) when using an EMG interface than when using a manual interface for high-frequency acceleration-based trajectory-tracking tasks. I also had participants with upper-arm disability after stroke perform the tracking task with the EMG and manual interfaces, and found that they adapted their feedforward controllers similarly to participants without disabilities for both interfaces. This suggests that EMG interfaces could enable accessible device interactions for people with disabilities if an assistive algorithm helps minimize errors arising from motor disturbances.
I then used game theory to model and enhance HMIs during continuous disturbance-rejection tasks. There are significant challenges with applying traditional control theory methods to quantify how humans and interfaces co-adapt to reference-tracking and disturbance-rejection tasks. Game theory provides a framework to model such two-learner problems. I modeled the human and the interface as two separate agents who are both trying to minimize task error and effort. Based on the results of my previous two aims, I developed an adaptive interface that augments the person’s feedback controller and found through simulation and experiment that co-adaptation improved performance and decreased human effort.
Lastly, I worked directly with people with disabilities to identify how technology can support people’s health and accessibility. I qualitatively assessed through an interview study how wearable sensors could support physical therapy access for people with disabilities to improve adherence and function. People's access to physical therapy was hampered by both social (e.g., physically visiting a clinic) and physiological (e.g., chronic pain) barriers. I defined core design principles (flexibility, movement tracking, community building) and tensions (insurance) to consider when developing technology to support physical therapy access.
Individualized HMIs with multi-channel, closed-loop control provides exciting opportunities for improving the accessibility of current and future technologies and improving health outcomes for people with and without disabilities. To advance our field's capacity to design and optimize interfaces that can adapt to individual users, both user-centered and quantitative approaches to enhancing interfaces must be developed. My dissertation provides the foundation for developing biosignal-based interfaces that support accessibility and health for people with and without disabilities.
