Thingamabobas Installation & HRI

• Thingamabobas Installation is a UK-based interactive art environment that combines robotic performance with kinetic sculpture and narrative storytelling to engage children.
• It utilizes an advanced sensor and vision system, including Intel RealSense cameras and YOLOv8, to dynamically transition between Sleep, Search, and Interact states based on audience proximity.
• Observational studies reveal both effective child engagement strategies and limitations in robotic expressivity, guiding future enhancements in human–robot interaction for performative installations.

The Thingamabobas Installation is a UK-based performative art environment that centers on facilitating novel, interactive encounters between children and a costumed robotic arm performer known as NED (Never-Ending Dancer). Conceived and evaluated as both an artistic and technological experiment, this installation blends kinetic sculpture, narrative framing, and real-time sensor-driven interaction. Rigorous observational studies have elucidated both the successes and challenges inherent in integrating autonomous robotic performers into child-facing cultural settings, with implications for the optimization of human–robot interaction (HRI) in interactive art.

1. System Architecture and Aesthetic Integration

At the heart of the Thingamabobas installation is NED—a 6-axis Niryo NED robotic arm transformed into a theatrical kinetic entity through costume, sound, and behavioral programming. The costume comprises long blue feathers affixed to the arm, imparting a bird-like silhouette, and a complex metallic skirt producing dynamic light reflections during operation. The robot’s blue base and joint surrounds create marked visual distinctiveness among the broader array of circus-themed sculptures populating the installation.

Functionally, NED’s interactive capabilities are dictated by an evolving sensor and vision pipeline: the system initially deployed an Intel RealSense D415 depth camera and subsequently adopted three D435 units to attain a panoramic 261° field-of-view. Pose and facial detection—implemented via models culminating with YOLOv8 integration—enable dynamic user localization and orientation, specifically tuned to detect and track children. The installation is further embedded in a cohesive performative environment utilizing a bespoke musical soundtrack and a prerecorded narrative from a “Thingamabobas Wrangler,” contextualizing the robot’s actions as part of an immersive, playful storyline.

2. Robotic Behavioral States and Transition Logic

NED’s behavioral repertoire is structured into three discrete, preprogrammed states—Sleep, Search, and Interact—activated according to real-time audience proximity. The logic governing these transitions is explicitly parameterized as follows:

State	Distance Domain	Behavior
Sleep	$\geq 4.0$ m	Arm folded ("resting swan"), inactivity, awaits stimuli
Search	$3.0 \leq d < 4.0$ m	Arm scanning, outward orientation, seeks engagement
Interact	$0.1 \leq d < 3.0$ m	Tracks and mimics user movement, encourages play

Transitions between these states are triggered exclusively by the detected user distance, as measured by the multi-camera configuration. In Sleep, NED presents minimal motion, serving both as a visual invitation to “wake” the robot and as a system-level inactivity mode. In Search, its posture becomes exploratory, aiming to draw in potential participants. Upon entering the Interact state, the robot employs real-time tracking to follow user gestures and positions, fostering an interactive mimicry—most frequently a form of dance.

3. Child–Robot Interaction: Observed Challenges

Empirical analysis of in-the-wild deployments revealed three principal obstacles constraining the depth and quality of child–robot engagement:

1. Initiating and Maintaining Engagement: Seventeen out of eighteen observed children utilized multimodal strategies (waving, arm spreading, verbal cues) to initiate interaction with NED, especially when interacting with a non-responsive or dormant robot. Successive maintenance strategies included performative mimicry and playful movement.
2. Lack of Expressivity and Reciprocity: Despite state-of-the-art tracking and imitation pipelines, NED's mechanical gestures lacked variability, resulting in responses that were perceived as monotone or insufficiently dancer-like. Emotional content codes such as "Confused," "Discouraged," and "Frustrated" were prevalent when children’s engagement strategies failed to elicit reciprocal or dynamically expressive responses.
3. Unmet Expectations: The robot’s designation as the “Never-Ending Dancer” cultivated substantial anticipation for continuous, lively behavior. Departure from this—whether through Sleep-induced inactivity or non-reciprocation of complex dance gestures—frequently triggered disengagement and negative emotional reactions in participants.

The root cause of these challenges included both the inherent limitations of current robotic expressivity and the misalignment between narrative/thematic cues and realized robotic capabilities.

4. Methodological Approaches to Observational Analysis

The empirical examination of child–robot interaction within the installation was conducted through a combination of video-based ethnography and qualitative interview analysis. Data acquisition encompassed over four hours of video footage from two venues, manually annotated using ELAN software to capture interaction modalities, gesture types, participant distance, and corresponding robot states. Immediate post-experience semi-structured interviews were conducted with each participant, transcribed, and coded thematically in NVivo.

Findings were structured into three principal themes: initiation and maintenance of engagement, expressivity/reciprocity limitations, and impact of unmet expectations. Notably, children’s attempts at engagement were rich and multimodal; however, recurring affective responses correlated strongly with robot state transitions and limitations in the expressive behavioral repertoire.

5. Strategies for Optimizing HRI in Performative Installations

Lessons from Thingamabobas directly inform strategies for improving HRI in similar contexts. Recommendations derived from the observed limitations include:

• Enhancing Expressivity: Expanding the motion vocabulary of performers like NED through motion pattern analysis and dynamic replay can foster a more convincing and engaging illusion of dancer-like autonomy and reciprocity. Fine-tuning algorithmic parameters to allow expressive, rather than merely mimetic, responses is critical.
• Revising State Design: The practicality and desirability of the Sleep state is questioned; modifications may include additional engagement triggers, such as voice recognition modules, to allow for more natural reactivation. Improved transition robustness can be achieved by deploying advanced pose estimation to minimize missed engagement opportunities.
• Expectation Management: Alignment of nomenclature, narrative framing, and actual system capabilities is necessary to mitigate dissonance between anticipatory cues and realized experience. The integration of multimodal input channels (gesture and voice) is cited as a pathway to richer, more meaningful interaction experiences.
• Technical Parameters: Adjustment of state boundaries (e.g., Sleep: $d\geq 4$ m; Search: $3 \leq d < 4$ m; Interact: $0.1 \leq d \leq 2.9$ m) and augmentation of response pattern granularity can be implemented within the existing software and sensing pipeline.

6. Impact and Implications for Young Audiences

Despite technical limitations, Thingamabobas was broadly successful in leveraging children’s innate curiosity and propensity for interactive experimentation. All child participants reported a positive overall experience, with seventeen expressing intent to revisit the installation. Observed behaviors demonstrated not only proficiency but also adaptability in exploring novel interaction modalities. Notably, the installation’s immersive narrative and strong aesthetic cues contributed to sustained engagement, even in the presence of robotic shortcomings.

This suggests that the integration of compelling artistic framing and multimodal stimuli can partially compensate for interactional limitations of current robotic technology. A plausible implication is that future performative installations may benefit from tightly coupled narrative, aesthetic, and technological components to maximize audience engagement and feedback quality for iterative HRI development.

7. Conclusions and Future Directions

The Thingamabobas installation exemplifies the interdisciplinary challenges and opportunities at the intersection of autonomous robotics and performative art. Empirical findings underscore the sensitivity of child–robot interaction to both technical expressivity and the narrative framing of robotic actors. Systematic improvement of behavioral variety, robustness of engagement triggers, and alignment of interactive capabilities with audience expectations are necessary for the advancement of meaningful HRI in similar artistic contexts. These insights inform not only the refinement of installations like Thingamabobas but also the broader development of interactive robotic systems aimed at public engagement and education.