Presentation
Designing Patient–Robot Communication Patterns in Hospitals: A Lean UX Research Approach
SessionPoster Session 2
DescriptionIntroduction:
Human-robot interaction (HRI) is a rapidly evolving field in healthcare, especially with their applications in hospital and clinical facilities, to support the staff and aid the patients navigate the environment. Hospital environments face increasing challenges due to staff shortages. For patients, the staff shortage leads to longer patient wait time, fragmented communications, and interruptions in care, often critical in circumstances where minutes can alter clinical outcomes. Large hospital campuses can also be confusing and physically taxing for patients and their visitors, exacerbating stress and delaying access to services. Early HRI applications, such as service robots for waiting-room support and transport robots for material supply benefit patients when they free staff and clinicians’ time for more critical bedside care.
A key question now is how patients and visitors perceive a robot’s communication both verbally (e.g., via word choice, tone, timing) and non-verbally (e.g., via gaze, gestures, lights). In busy, high-stress hospitals, these cues affect clarity, responsiveness, and efficiency. For example, clear phrasing of the robots with directional gestures helps people quickly understand what to do and where to go. Adaptive behaviors, like slowing when someone lags, make the robot responsive to changing needs. And also explicit handoff confirmations reduce back-and-forth, shorten tasks, and lead to a better patient experience. Importantly, the impact of these communication patterns depends on the scenario, such as a transportation robot versus a service robot in emergency or service contexts.
Therefore, this study explores how patients and visitors perceive robots’ verbal and non-verbal signals across different hospital scenarios, with the goal of improving clarity, responsiveness, and efficiency in HRI. A lean UX research approach is essential here because it enables early and efficient testing of communication patterns in real-world hospital contexts. In these settings, human–robot collaboration must be carefully designed so that robots complement rather than complicate staff workflows, reduce rather than add to patient stress, and adapt fluidly to dynamic hospital demands. This iterative, low-cost validation helps confirm collaborative interactions before larger-scale deployment.
Approach:
We approach the problem through the Lean UX research approach, which emphasizes rapid, iterative testing with real users in naturalistic settings. To investigate how patients and caregivers perceive robots in hospitals and how they choose to engage with the robots, five scenarios were pre-defined to simulate realistic hospital interactions.
1. Wayfinding to speciality departments (e.g. cardiology): The service robot should confidently lead or escort patients, confirm the route and handoff the patient to the care team.
2. Emergency state guidance (fire drill or real situation): Clearly communicate cues to guide small groups of people, reporting headcount at the safety zone before staff handoff.
3. Visitor assistance to specific units (finding ICU or maternity ward): Escort or lead the visitors with simple visual cues and hand off the visitors at the desk.
4. Service support (waiting time, visit history, appointment details): Display initials with wait time and ticket ID on screen, offer a QR code to send more details on personal devices.
5. Equipment/scrub transportation (no direct interaction): Clearly display “not seeking interaction" and biohazard visually.
For phase 1, we employed semi-structured, artifact assisted interviews (n = 5) using storyboards and prompts to elicit user expectations for robot behavior in each scenario. Participants (patient and visitor roles) are walked through each scenario, describing how they would communicate with the robot or seek help using storyboards and short prompts (wayfinding, emergency guidance, visitor assistance, service support, transportation robot awareness). The moderator followed a brief script (outline: discoverability, seeking interaction, communication cues from robots, hand-offs and privacy). The participants were prompted to describe what they would do, expect or say in each scenario. All the data was captured through notes and in-person audio recording via hand-held recording device. The primary dataset comprises moderator notes and participant responses.
Findings:
In the pilot study conducted, five adults (n=5) were a mixed sample of visitors and patients (ages ≈20s–30s; education ranging undergraduate to graduate).
● Scenario 1: 100% of participants generally preferred and were open to robot-led wayfinding across multiple hospital departments. Participants indicated that the communication formats for this task should be primarily non-verbal, relying on light cues to signal direction. However, verbal communication was expected in specific contexts, such as during hand-offs to the care team, where clarity and confirmation were seen as essential.
● Scenario 2: 80% of participants reported low trust in robot-led emergency guidance, although 40% expressed that they were conditionally comfortable with a small group led to the safety area by the robots. They preferred the robot to clearly communicate cues for guidance, next steps, audio directions and reported a headcount to the rescue team. The communication format was expected as primarily non-verbal for directional guidance, supplemented by verbal communication for staff hand-offs.
● Scenario 3: 60% of participants preferred visitor drop-off or hand-off at the desk when prompted. The communication format was expected to be non-verbal with light cues for directional guidance.
● Scenario 4: All participants expressed their discomfort in the robot saying their PHI out loud. 40% of participants explicitly express being more comfortable with a token number shared by the robot, with a QR code for next steps, including access to their care team that can be viewed through their personal devices. In this context, the preferred communication format emphasized non-verbal methods and screen-based identifiers, while audio cues were not expected.
● Scenario 5: All participants expected the transportation robots to display unambiguous cues stating “not seeking interaction” and bio-hazard labels if applicable. For this scenario, the preferred communication format non-verbal only, with visual cue like labels expected.
● Coss-cutting probes: The participants were then asked by what conditions would they approach the robot for help. 40% of participants mentioned that they were likely to engage a robot for assistance only if no humans were present nearby. However, their trust rose when they were inquired if the robots would approach them proactively and had clear multi-modal communication cues (visual and audio).
Takeaways:
A lean research approach proved fast, low-cost and non-invasive for users in eliciting trust, privacy and their expected modalities of communication. Storyboarding and modality tokens gave the participants a visual representation of the real-life scenarios and clarified trade-offs without deploying robots and disrupting the hospital environment. Phase 1 revealed specific expectations in hand-offs, communication, privacy-by-default PHI handling, clear communication of the robot seeking interaction, as well as human intervention. Five participants were recruited as a pilot sample for early interaction pattern discovery and prioritized design specifications and formed the foundations for phase 2 hypotheses.
Future Experimentation:
In Phase 2, we will expand the study to 25–30 participants, ensuring representativeness across age, education levels, cultural background, and access needs (such as low vision or mobility constraints). Each 45–60 minute session will involve artifact-assisted, semi-structured interviews with a participant mixed role of patients and visitors. Updated storyboards will be used to restructure scenario prompts, aiming to surface interaction-seeking behaviors and expectations, while systematically exploring modality tokens such as lighting, motion, voice, and display. In addition to probing user preferences and trust, this phase will begin to survey the mapping between robot type, communication format (verbal/non-verbal), and application scenario, as this relationship is critical to designing effective human–robot collaboration in hospitals. During the sessions, micro-measures of trust, clarity of interaction, and privacy/comfort will be noted. Analysis will include matrix coding (scenario × modality × role) to identify cross-scenario patterns, triangulation of Likert ratings for negative-case analysis to surface dissenting views, and saturation tracking using a scenario × theme saturation grid.
Human-robot interaction (HRI) is a rapidly evolving field in healthcare, especially with their applications in hospital and clinical facilities, to support the staff and aid the patients navigate the environment. Hospital environments face increasing challenges due to staff shortages. For patients, the staff shortage leads to longer patient wait time, fragmented communications, and interruptions in care, often critical in circumstances where minutes can alter clinical outcomes. Large hospital campuses can also be confusing and physically taxing for patients and their visitors, exacerbating stress and delaying access to services. Early HRI applications, such as service robots for waiting-room support and transport robots for material supply benefit patients when they free staff and clinicians’ time for more critical bedside care.
A key question now is how patients and visitors perceive a robot’s communication both verbally (e.g., via word choice, tone, timing) and non-verbally (e.g., via gaze, gestures, lights). In busy, high-stress hospitals, these cues affect clarity, responsiveness, and efficiency. For example, clear phrasing of the robots with directional gestures helps people quickly understand what to do and where to go. Adaptive behaviors, like slowing when someone lags, make the robot responsive to changing needs. And also explicit handoff confirmations reduce back-and-forth, shorten tasks, and lead to a better patient experience. Importantly, the impact of these communication patterns depends on the scenario, such as a transportation robot versus a service robot in emergency or service contexts.
Therefore, this study explores how patients and visitors perceive robots’ verbal and non-verbal signals across different hospital scenarios, with the goal of improving clarity, responsiveness, and efficiency in HRI. A lean UX research approach is essential here because it enables early and efficient testing of communication patterns in real-world hospital contexts. In these settings, human–robot collaboration must be carefully designed so that robots complement rather than complicate staff workflows, reduce rather than add to patient stress, and adapt fluidly to dynamic hospital demands. This iterative, low-cost validation helps confirm collaborative interactions before larger-scale deployment.
Approach:
We approach the problem through the Lean UX research approach, which emphasizes rapid, iterative testing with real users in naturalistic settings. To investigate how patients and caregivers perceive robots in hospitals and how they choose to engage with the robots, five scenarios were pre-defined to simulate realistic hospital interactions.
1. Wayfinding to speciality departments (e.g. cardiology): The service robot should confidently lead or escort patients, confirm the route and handoff the patient to the care team.
2. Emergency state guidance (fire drill or real situation): Clearly communicate cues to guide small groups of people, reporting headcount at the safety zone before staff handoff.
3. Visitor assistance to specific units (finding ICU or maternity ward): Escort or lead the visitors with simple visual cues and hand off the visitors at the desk.
4. Service support (waiting time, visit history, appointment details): Display initials with wait time and ticket ID on screen, offer a QR code to send more details on personal devices.
5. Equipment/scrub transportation (no direct interaction): Clearly display “not seeking interaction" and biohazard visually.
For phase 1, we employed semi-structured, artifact assisted interviews (n = 5) using storyboards and prompts to elicit user expectations for robot behavior in each scenario. Participants (patient and visitor roles) are walked through each scenario, describing how they would communicate with the robot or seek help using storyboards and short prompts (wayfinding, emergency guidance, visitor assistance, service support, transportation robot awareness). The moderator followed a brief script (outline: discoverability, seeking interaction, communication cues from robots, hand-offs and privacy). The participants were prompted to describe what they would do, expect or say in each scenario. All the data was captured through notes and in-person audio recording via hand-held recording device. The primary dataset comprises moderator notes and participant responses.
Findings:
In the pilot study conducted, five adults (n=5) were a mixed sample of visitors and patients (ages ≈20s–30s; education ranging undergraduate to graduate).
● Scenario 1: 100% of participants generally preferred and were open to robot-led wayfinding across multiple hospital departments. Participants indicated that the communication formats for this task should be primarily non-verbal, relying on light cues to signal direction. However, verbal communication was expected in specific contexts, such as during hand-offs to the care team, where clarity and confirmation were seen as essential.
● Scenario 2: 80% of participants reported low trust in robot-led emergency guidance, although 40% expressed that they were conditionally comfortable with a small group led to the safety area by the robots. They preferred the robot to clearly communicate cues for guidance, next steps, audio directions and reported a headcount to the rescue team. The communication format was expected as primarily non-verbal for directional guidance, supplemented by verbal communication for staff hand-offs.
● Scenario 3: 60% of participants preferred visitor drop-off or hand-off at the desk when prompted. The communication format was expected to be non-verbal with light cues for directional guidance.
● Scenario 4: All participants expressed their discomfort in the robot saying their PHI out loud. 40% of participants explicitly express being more comfortable with a token number shared by the robot, with a QR code for next steps, including access to their care team that can be viewed through their personal devices. In this context, the preferred communication format emphasized non-verbal methods and screen-based identifiers, while audio cues were not expected.
● Scenario 5: All participants expected the transportation robots to display unambiguous cues stating “not seeking interaction” and bio-hazard labels if applicable. For this scenario, the preferred communication format non-verbal only, with visual cue like labels expected.
● Coss-cutting probes: The participants were then asked by what conditions would they approach the robot for help. 40% of participants mentioned that they were likely to engage a robot for assistance only if no humans were present nearby. However, their trust rose when they were inquired if the robots would approach them proactively and had clear multi-modal communication cues (visual and audio).
Takeaways:
A lean research approach proved fast, low-cost and non-invasive for users in eliciting trust, privacy and their expected modalities of communication. Storyboarding and modality tokens gave the participants a visual representation of the real-life scenarios and clarified trade-offs without deploying robots and disrupting the hospital environment. Phase 1 revealed specific expectations in hand-offs, communication, privacy-by-default PHI handling, clear communication of the robot seeking interaction, as well as human intervention. Five participants were recruited as a pilot sample for early interaction pattern discovery and prioritized design specifications and formed the foundations for phase 2 hypotheses.
Future Experimentation:
In Phase 2, we will expand the study to 25–30 participants, ensuring representativeness across age, education levels, cultural background, and access needs (such as low vision or mobility constraints). Each 45–60 minute session will involve artifact-assisted, semi-structured interviews with a participant mixed role of patients and visitors. Updated storyboards will be used to restructure scenario prompts, aiming to surface interaction-seeking behaviors and expectations, while systematically exploring modality tokens such as lighting, motion, voice, and display. In addition to probing user preferences and trust, this phase will begin to survey the mapping between robot type, communication format (verbal/non-verbal), and application scenario, as this relationship is critical to designing effective human–robot collaboration in hospitals. During the sessions, micro-measures of trust, clarity of interaction, and privacy/comfort will be noted. Analysis will include matrix coding (scenario × modality × role) to identify cross-scenario patterns, triangulation of Likert ratings for negative-case analysis to surface dissenting views, and saturation tracking using a scenario × theme saturation grid.
Event Type
Poster Presentation
TimeTuesday, March 244:45pm - 6:15pm EDT
LocationRhinelander Gallery
Hospital Environments