Peek into the `White-Box': A Field Study on Bystander Engagement with Urban Robot Uncertainty

Research Background and Issues

Issues and Challenges

• Challenges of Robot Uncertainty: Robots inherently face uncertainty in autonomous decision-making due to factors such as sensor noise and the limitations of machine learning models. This issue becomes particularly pronounced in dynamic and ever-changing public urban spaces. For instance, a widely discussed incident in 2022 involved a delivery robot crossing a police cordon and entering a crime scene, exposing the imperfections of robotic technology.
• Lack of Human Assistance for Robots in Urban Environments: In public spaces, robots often need to handle uncertainties independently without direct collaborators, which can lead to operational errors and affect public acceptance.

Importance

• Collaborating with humans to address robot uncertainty not only helps optimize robotic decision-making but also fosters trust and interaction between humans and robots.
• Introducing robots into urban environments requires addressing public concerns about safety, reliability, and potential societal impacts (e.g., invisible labor or job displacement).

Research Motivation and Related Work

• Motivation: To explore how to engage bystanders in public spaces in a non-intrusive manner to help robots address uncertainty, thereby enhancing public understanding and acceptance of robots.
• Related Work:
  • Research on human-robot interaction often focuses on users as explicit collaborators, while studies on "informal collaboration" between bystanders and robots remain limited.
  • Existing studies on "incremental machine learning" and "dynamic non-verbal interaction strategies" provide a foundation for the discussions in this paper.

Solution

Approach or Solution

• A "peephole" design concept is proposed to support bystanders in assisting robots with urban uncertainties. Inspired by the "black box" AI metaphor, this design employs a curiosity-driven and non-intrusive approach to present information about the robot's need for assistance.

Innovations of the Solution

1. Non-intrusiveness: Avoids direct verbal requests by subtly displaying information (e.g., through a peephole) to stimulate bystanders' initiative.
2. Curiosity Stimulation: Utilizes "peeping" as an interaction mode, transforming ordinary help requests into exploratory, gamified interactions.
3. Enhanced Emotional Interaction: The method is based on "gamification logic in human-robot interaction," attracting bystanders' attention through the concealment and revelation of information.

Implementation Steps and Key Technologies

1. Hardware Implementation:
   • A mobile robot modeled after street delivery robots, equipped with a flip-open binocular device.
   • During the flipping process, the robot displays "uncertainty information" encountered in environmental recognition, including problem descriptions and help requests.
2. Design Enhancements:
   • Added gesture cues (an animated "flagpole" that sways and waves to attract attention) and audio cues (beeping sounds).
3. Wizard-of-Oz Experimentation: Two researchers remotely controlled the robot to simulate uncertain scenarios (e.g., encountering puddles or leaves) and collected bystander reactions.

Research Outcomes

Specific Findings

1. Collected natural behavioral and interaction experience data from bystanders helping the robot resolve uncertainties on-site.
2. Found that "curiosity" and "human initiative" drove bystanders to assist the robot. After the interaction, participants exhibited positive emotional responses, such as laughter and surprise.
3. Revealed how participation in helping behaviors can enhance bystanders' trust and emotional connection with the robot.

Comparison with Existing Solutions and Advantages

1. Compared to explicit methods of requesting human assistance, the peephole design leverages bystanders' curiosity and exploratory interest, reducing operational risks and burdens.
2. Strengthened public perceptions of "safety" and "human superiority," partially alleviating anxieties about full robotic automation.
3. The robot's display of "imperfection" was perceived as "cute and endearing," aligning with human traits and helping to build emotional bonds between humans and robots.

Experimental and Evaluation Results

• Engagement: Among the crowd that first noticed the robot, only 29 instances of actual assistance occurred (relative to the low foot traffic), but only two bystanders who initiated interaction dropped out midway, indicating a high completion rate.
• Interaction Effectiveness:
  • Participants generally gave positive feedback on the peephole design.
  • Some individuals felt empowered and responsible after helping the robot, leading to increased trust in the robot.
  • The combination of animated flagpole and audio cues significantly improved understanding and willingness to assist.

Limitations and Future Directions

1. Limitations:
   • The study environment was limited to a university campus, where bystanders' educational backgrounds and open-mindedness may differ from those in other public settings.
   • The dynamic evolution of the experimental design increased data complexity, leading to inconsistent results across different scenarios.
2. Future Directions:
   • Explore how to implement the peephole concept in diverse public settings.
   • Investigate how robots can more clearly convey their intentions in help-seeking scenarios to eliminate participation barriers.
   • Further explore strategies such as gamification, emotion-driven approaches, or social influence to promote spontaneous collaboration between humans and robots.