Designing Robots for Families: In-Situ Prototyping for Contextual Reminders on Family Routines

1. Problem Statement

While social robots are increasingly entering domestic environments, their long-term integration into family life remains a significant challenge, often leading to rapid disengagement after initial novelty wears off.

• The Gap: Existing reminder systems are often time-based and lack context, making them insufficient for complex family routines (e.g., medication adherence or chore completion).
• The Opportunity: Robots possess unique capabilities (mobility, sensing, social presence) to leverage rich contextual data (time, location, environment state, and specific family members) to provide proactive, situationally relevant reminders.
• The Challenge: Designing robots that navigate the fluid, dynamic, and socially complex nature of household routines without causing friction, privacy concerns, or authority conflicts.

2. Methodology

The authors employed a Research through Design (RtD) approach, utilizing a two-phase process involving 10 families (with children aged 7–14).

Phase 1: Co-Design Sessions (n=10)

• Goal: To understand existing family routines and co-design robot behaviors.
• Process:
  • Families selected ~5 routines from a pre-defined list (based on the Family-Robot Routines Inventory).
  • Bodystorming: Participants imagined acting as the robot to define trigger conditions and actions.
  • Mapping: Researchers mapped key household locations using the robot's navigation system.
• Outcome: Identification of four critical dimensions for contextual reminders:
  1. Time: Specific ranges (not just fixed points) to balance missed opportunities vs. irrelevant interruptions.
  2. Location: Specific zones where triggers occur.
  3. Family Members: Target recipients (usually children, sometimes the whole family).
  4. State of Environment: Visual or activity-based cues (e.g., "child is holding a tablet" or "toys are scattered").

Phase 2: In-Situ User Study (n=8)

• Deployment: A 4-day in-home study with a mobile social robot prototype.
• Robot Platform: Temi Robot (mobile, screen-integrated) controlled by a Raspberry Pi 5.
• System Architecture:
  • Scheduler: Determines robot location and attention based on the customized routine plan.
  • Responder: Event-driven module handling state updates and communication.
  • Image Analyzer: A two-step visual processing pipeline:
    1. YOLO11: Lightweight object detection (people/objects).
    2. Vision-Language Model (VLM - GPT-4.1): High-level scene understanding to confirm context (e.g., "Are they doing homework together?").
  • Interaction: The robot uses a wake word ("Hey, Temi") and a cloud-based LLM (GPT-4.1-mini) for conversational responses and embedding reminders into dialogue.
• Data Collection: Semi-structured interviews (parents and children), system logs, conversation transcripts, and visual snapshots.

3. Key Contributions

1. Design Space Definition: The paper outlines a structured design space for robot-facilitated reminders, defining the four core contextual dimensions (Time, Location, Members, Environment State) and auxiliary constraints (frequency, persistence).
2. Technical Artifact: Development and deployment of a functional mobile robot system capable of autonomous navigation, visual context sensing (YOLO + VLM), and proactive reminder delivery.
3. Empirical Insights: A deep understanding of how families perceive and interact with contextual reminders, highlighting the shift from "time-based" to "context-aware" interactions.
4. Practical Design Implications: Actionable guidelines for future HRI in domestic settings, focusing on the balance between autonomy and family authority.

4. Results and Findings

Perceived Impact on Routines

• Positive Reception: Families found the robot helpful, particularly as a neutral communicator. Parents appreciated offloading the "nagging" burden, noting that children were more receptive to reminders from a robot than from parents.
• Effectiveness: The robot successfully facilitated timely task completion (e.g., brushing teeth, homework).
• Limitations: Reminders are not enforceable. If a child chose to ignore the robot, the reminder failed. The robot lacked the authority to compel action, leading to situations where children evaded detection (e.g., moving to a different room).

User Experience (UX)

• Enjoyment: Families enjoyed the robot's mobility, conversational abilities (jokes, facts), and its emergence as a "family member."
• Technical Friction: Navigation failures were a major pain point. Issues included:
  • Inability to traverse carpets or multi-level homes.
  • Failure to dock/charge when environments changed (e.g., moved furniture).
  • Silent Failures: The robot often failed to charge or navigate without alerting the family, leading to downtime.
• Contextual Errors: The system occasionally misidentified contexts (e.g., mistaking reading a book for screen time, or misidentifying a child as a parent). However, families generally found these errors humorous rather than frustrating.

Family-Robot Dynamics

• Authority Tensions: The robot's role as a reminder agent sometimes conflicted with family hierarchies. While it reduced parental stress, it could not replace parental authority in enforcement.
• Privacy: Concerns were minimal due to the robot's restricted movement and local processing, though parents utilized "privacy mode" for sensitive work areas.

5. Significance and Design Implications

The study concludes that integrating robots into families is an ecosystemic process where routines shape the robot, and the robot reshapes the routines.

Key Design Implications:

• Prioritize Proactivity: Families prefer proactive nudges over manual checks, as they serve as necessary "pushes" for tasks users might otherwise ignore.
• Loop in Parents: To address the lack of enforcement, robots should be designed to loop in parents (e.g., notifying them of non-compliance) rather than acting as the sole authority.
• Help-Seeking Behavior: Robots must be designed to explicitly ask for help when they fail (e.g., navigation errors) rather than failing silently.
• Hybrid Coverage: Given the physical limitations of mobile robots, a "sidekick" approach (combining mobile robots with static sensors or screens) may be necessary for full home coverage.

Ethical Considerations:
The paper emphasizes the need for Human-Centered AI (HCAI) principles, including transparent data handling, user-controlled privacy switches, and the ethical responsibility of adult decision-makers regarding children's privacy and safety in the presence of social robots.

Conclusion:
This work demonstrates that while contextual reminders are a promising entry point for robot integration, successful long-term adoption requires addressing technical robustness (navigation/sensing), social dynamics (authority/enforcement), and ethical transparency.