Humanoid robots are increasingly expected to be designed to interact in diverse social contexts. Yet, recent commercial prototypes often omit expressive communication features, such as faces and gestures. This disconnect raises questions about how communication modalities are actually selected for real-world deployment. In this study, we interviewed 12 industry decision-makers involved in humanoid robot development to examine the factors that guide their design adoption choices. Our findings show that technical feasibility, cost, safety, reliability, and organizational priorities frequently outweigh the benefits of communication modalities. By highlighting where practitioner considerations align with or diverge from established design research, this study offers a grounded, industry-facing perspective on the design of humanoid communication. Interpreting our findings, we present a conceptual framework to guide design researchers in understanding decision-making for product development in industry contexts.

Wizard-of-Oz (WoZ) is one of the most widely used experimental methodologies across the field of Human-Robot Interaction (HRI), making WoZ teleoperation interfaces a critical tool for HRI research. Yet current WoZ teleoperation interfaces are overwhelmingly tailored towards a narrow set of HRI interaction paradigms. In this work, we conducted a set of interviews with HRI researchers to better understand the diversity of teleoperation needs across the HRI community. Our analysis highlighted (1) human challenges, with respect to wizards’ expertise, the need for quick responses, and research participants’ unpredictability; (2) robot challenges, with respect to robot malfunctions, delays, and robot-driven complexity, and (3) interaction challenges, with respect to researchers’ varying control requirements and the need for precise experimental control. Moreover, our results revealed unexpected parallels between the experiences of HRI researchers and real-world teleoperators, which open up fundamentally new possibilities for future work in robot control interfaces and encourage radically different perspectives on what types of interfaces are even needed to best facilitate WoZ experimentation. Leveraging these insights, we recommend that WoZ interfaces (1) be designed with extensibility and customization in mind, (2) ease interaction management by accounting for unpredictability and multi-robot interactions, and (3) consider WoZ teleoperators beyond the context of experimentation.

Recent work on Socially Assistive Robotics in Therapy has revealed a dual-cycle model, with the vast majority of prior work on Socially Assistive Robotics narrowly focused on the human-robot interaction, termed the "inner cycle". In contrast, little attention has been paid to the activities performed before and after the interaction, termed the "outer cycle", in which authoring and evaluation also take place. Authoring and evaluation are activities that are key sources of invisible labor for Therapists who serve as Care Wizards (i.e., SAR teleoperators). In this work, we consider the outer cycle needs of Care Wizards in another key Socially Assistive Robotics domain, Special Education, with a careful eye toward the barriers to entry and invisible labor that may manifest in this domain, and how those barriers and invisible labor might be subverted and mitigated. Our interviews with six Care Wizards who teleoperate robots in Special Education contexts reveal new insights surrounding these stakeholders’ needs. Our key insights are that (1) support systems are necessary for SAR adoption; (2) currently invisible Care Wizard labor may be indirectly compensated; and (3) training must be personalized to specific Care Wizards.

Some labor is overlooked or devalued, while necessary within the context of paid employment. This is “invisible labor”. Invisible labor is often performed by minoritized groups and is typically invisible to those in power. Novel technologies can introduce new sociotechnical labor paradigms that reduce labor visibility. In this paper, we consider how invisible labor might manifest for teleoperated Socially Assistive Robots (SARs). By combining an analysis of the labor context of teleoperated SAR use with insights from interviews with SAR teleoperators, we demonstrate how invisible labor manifests in the practical deployment of teleoperated SARs. Finally, we provide recommendations for developers and policymakers to remedy this labor invisibility.

Socially assistive robots play an effective role in children's therapy and education. Robots engage children and provide interaction that is free of the potential judgment of human peers and adults. Research in socially assistive robots for children generally focuses on therapeutic and educational outcomes for those children, informed by a vision of autonomous robots. This perspective ignores therapists and educators, who operate these robots in practice. Through nine interviews with individuals who have used robots to deliver socially assistive services to neurodivergent children, we (1) define a dual-cycle model of therapy that helps capture the domain expert view of therapy, (2) identify six core themes of teleoperator needs and patterns across these themes, (3) provide high-level guidelines and detailed recommendations for designing teleoperated socially assistive robot systems, and (4) outline a vision of robot-assisted therapy informed by these guidelines and recommendations that centers teleoperators of socially assistive robots in practice.

Socially Assistive Robots are effective at supporting autistic children in a variety of different therapies. Therapists can control the robots' motions and verbalizations to engage children and deliver therapeutic interventions based on their needs. We present teleoperation capabilities to support therapists in customizing therapy to their clients' needs. Specifically, we introduce a documentation sidebar that aims to prime therapists using their clients' documented needs, and a session summary report that helps therapists reflect on the session with the child. We present preliminary designs for these capabilities and describe future work to build upon them.

Socially Assistive Robots (SARs) show significant promise in a number of domains: providing support for the elderly, assisting in education, and aiding in therapy. Perhaps unsurprisingly, SAR research has traditionally focused on providing evidence for this potential. In this paper, we argue that this focus has led to a lack of critical reflection on the appropriate level of autonomy (LoA) for SARs, which has in turn led to blind spots in the research literature. Through an analysis of the past five years of HRI literature, we demonstrate that SAR researchers are overwhelmingly developing and envisioning autonomous robots. Critically, researchers do not include a rationale for their choice in LoA, making it difficult to determine their motivation for fully autonomous robots. We argue that defaulting to research fully autonomous robots is potentially short-sighted, as applying LoA selection guidelines to many SAR domains would seem to warrant levels of autonomy that are closer to teleoperation. We moreover argue that this is an especially critical oversight as teleoperated robots warrant different evaluation metrics than do autonomous robots since teleoperated robots introduce an additional user, the teleoperator. Taken together, this suggests a mismatch between LoA selection guidelines and the vision of SAR autonomy found in the literature. Based on this mismatch, we argue that the next five years of SAR research should be characterized by a shift in focus towards teleoperation and teleoperators.

Socially Assistive Robots (SARs) have shown promise, but there are still practical challenges to their widespread adoption. Recent research has demonstrated the advantages of teleoperated systems in this space and called for better guidelines for teleoperation interfaces. We ran group usability tests with therapists with no experience with robots to learn more about the challenges they face. We found that robot-novice therapists understand how robots can be effective in therapy. However, learning to use a robot interface can be challenging for new users. These challenges include the unfamiliar metaphors used for robot connection and the need to create, acquire, or share robot interaction content. We also identify users’ needs that are perhaps non-obvious in a research context, such as privacy of client health information and professional boundaries with client families when using electronic tools. We make several recommendations based on analysis of our group usability tests: (1) developing dedicated interfaces for content authoring that account for caregiver technical expertise, (2) implementing content organization and sharing tools, (3) using connection metaphors that non-technical users may be more familiar with such as phone calls or web URLs, (4) considering user privacy in connection methods chosen, especially within telehealth. Most importantly, we encourage further research in SAR teleoperation that focuses on caregivers as teleoperators.

Socially assistive robots (SARs) receive a lot of research attention due to their positive impact in a variety of contexts. Importantly, studies have shown that children with autism are much more receptive to SARs in therapy while resulting in similar learning outcomes. Given the sensitive nature of therapy and the current state of autonomous robots, in practice robots are teleoperated by a therapist controlling their motion and dialogue. There is an opportunity to produce more effective teleoperation interfaces of SARs in the context of therapy for children with autism. In this paper, I outline research for improving teleoperation interfaces of SARs through (1) analyzing current teleoperation usage, (2) interviewing therapists about their needs, and (3) implementing and evaluating varied designs for teleoperation interfaces.

Therapist-operated robots can play a uniquely impactful role in helping children with autism practice and acquire social skills. While extensive research within Human-Robot Interaction has focused on teleoperation interfaces for robots in general, little work has been done on teleoperation interface design for robots in the context of therapy for children with autism. Moreover, while clinical research has shown the positive impact robots can have on children with autism, much of that research has been performed in a controlled environment, with little understanding of the way these robots are used in practice. We analyze archival data of therapists teleoperating robots as part of their regular therapy sessions, to (1) determine common themes and difficulties in therapists’ use of teleoperation interfaces, and (2) provide design recommendations to improve therapists’ overall experience. We believe that following these recommendations will help maximize the effectiveness of therapy for children with autism when using Socially Assistive Robotics and the scale at which robots can be deployed in this domain.

A critical challenge in using longitudinal neuroimaging data to study the progressions of Alzheimer's Disease (AD) is the varied number of missing records of the patients during the course when AD develops. To tackle this problem, in this paper we propose a novel formulation to learn an enriched representation with fixed length for imaging biomarkers, which aims to simultaneously capture the information conveyed by both baseline neuroimaging record and progressive variations characterized by varied counts of available follow-up records over time. Because the learned biomarker representations are a set of fixedlength vectors, they can be readily used by traditional machine learning models to study AD developments. Take into account that the missing brain scans are not aligned in terms of time in a studied cohort, we develop a new objective that maximizes the ratio of the summations of a number of ℓ1-norm distances for improved robustness, which, though, is difficult to efficiently solve in general. Thus, we derive a new efficient and non-greedy iterative solution algorithm and rigorously prove its convergence. We have performed extensive experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. A clear performance gain has been achieved in predicting ten different cognitive scores when we compare the original baseline biomarker representations against the learned representations with longitudinal enrichments. We further observe that the top selected biomarkers by our new method are in accordance with known knowledge in AD studies. These promising results have demonstrated improved performances of our new method that validate its effectiveness.

Alzheimer's Disease (AD) is a chronic neurodegenerative disease that severely impacts patients' thinking, memory and behavior. To aid automatic AD diagnoses, many longitudinal learning models have been proposed to predict clinical outcomes and/or disease status, which, though, often fail to consider missing temporal phenotypic records of the patients that can convey valuable information of AD progressions. Another challenge in AD studies is how to integrate heterogeneous genotypic and phenotypic biomarkers to improve diagnosis prediction. To cope with these challenges, in this paper we propose a longitudinal multi-modal method to learn enriched genotypic and phenotypic biomarker representations in the format of fixed-length vectors that can simultaneously capture the baseline neuroimaging measurements of the entire dataset and progressive variations of the varied counts of follow-up measurements over time of every participant from different biomarker sources. The learned global and local projections are aligned by a soft constraint and the structured-sparsity norm is used to uncover the multi-modal structure of heterogeneous biomarker measurements. While the proposed objective is clearly motivated to characterize the progressive information of AD developments, it is a nonsmooth objective that is difficult to efficiently optimize in general. Thus, we derive an efficient iterative algorithm, whose convergence is rigorously guaranteed in mathematics. We have conducted extensive experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) data using one genotypic and two phenotypic biomarkers. Empirical results have demonstrated that the learned enriched biomarker representations are more effective in predicting the outcomes of various cognitive assessments. Moreover, our model has successfully identified disease-relevant biomarkers supported by existing medical findings that additionally warrant the correctness of our method from the clinical perspective.

With rapid progress in high-throughput genotyping and neuroimaging, researches of complex brain disorders, such as Alzheimer's Disease (AD), have gained significant attention in recent years. Many prediction models have been studied to relate neuroimaging measures to cognitive status over the progressions when these disease develops. Missing data is one of the biggest challenge in accurate cognitive score prediction of subjects in longitudinal neuroimaging studies. To tackle this problem, in this paper we propose a novel formulation to learn an enriched representation for imaging biomarkers that can simultaneously capture both the information conveyed by baseline neuroimaging records and that by progressive variations of varied counts of available follow-up records over time. While the numbers of the brain scans of the participants vary, the learned biomarker representation for every participant is a fixed-length vector, which enable us to use traditional learning models to study AD developments. Our new objective is formulated to maximize the ratio of the summations of a number of ℓ1-norm distances for improved robustness, which, though, is difficult to efficiently solve in general. Thus we derive a new efficient iterative solution algorithm and rigorously prove its convergence. We have performed extensive experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. A performance gain has been achieved to predict four different cognitive scores, when we compare the original baseline representations against the learned representations with enrichments. These promising empirical results have demonstrated improved performances of our new method that validate its effectiveness.