Designing effective Human-Robot Interaction in task-oriented settings requires carefully balancing user engagement with socially acceptable levels of robot intrusiveness. In this paper, we examine how different robot intervention strategies shape user experience, interaction dynamics, perceived intrusiveness, and sense of support. We compare two approaches: a continuous engagement-seeking robot strategy, and a context-aware strategy that selectively intervenes based on the user's state and task context. Both approaches rely on multimodal behavioral cues, including body orientation and attentional signals, to guide robot actions. We evaluate these strategies in a user study with 32 participants performing a task in a simulated hospital environment. Our findings show that higher interaction frequency does not necessarily lead to better engagement. Instead, we observe a systematic trade-off between perceived support and intrusiveness, influenced by factors such as physical proximity and user effort. These results provide empirical evidence that effective engagement in HRI depends on adaptive, context-sensitive intervention policies.
Figures & tables
Fig. 1 : Context-aware robot behavior. Top : Participant focused on the task, showing no intention to interact with the robot. Bottom : RViz visualization of the robot’s perception.
Fig. 2 : Experimental setup. Top : Participant facing the robot, indicating an intention to interact. Bottom : RViz visualization of the robot’s perception. The yellow bounding box marks the person, while the red and blue arrows indicate the torso and head directions, respectively.
Fig. 3 : Proposed systems. Left : continuous engagement system. Right : balanced controller system. Both systems use visual input to track the person. In System 1, a continuous controller drives motion and dialogue, with the LLM invoked only under specific conditions. In System 2, the LLM controls responses and motion based on user input and visual context, with a balanced controller triggering occasional interventions when no audio input is detected.
Fig. 4 : Robot and task items. (a) Robot screen showing listening (green) and responding (blue) states, along with dialog transcripts; a red skip button interrupts responses and resumes listening. (b) Food, (c) Fluids, (d) Medication pills.
System
S1
S2
Duration (s)
303.2±91.5
342.3±113.3
Position changes
7.6±3.1
11.3±5.5
User Inputs
18.3±9.9
21.8±10.4
System Outputs
28.1±10.5
25.8±11.5
TABLE I : Experiment metrics . Average values measured.
Fig. 5 : Correlations between objective and subjective metrics. Scatter plots showcasing correlation for both systems, grouped by objective metric. Dashed lines and solid lines indicate non-significant and significant correlations ( p<0.05 ), respectively. Legend with statistics’ values: Valence Score (VS), Arousal Score (AS), Reward Factor (RF), Perceived Intelligence (PI), and Perceived Usability (PU).