Abstract
This study explores how human perceptions of a non-anthropomorphic robotic manipulator can be shaped by two key dimensions of behaviour: arousal, defined as the robot's movement energy and expressiveness, and attention, defined as the robot's capacity to selectively orient toward and engage with a user. We present an integrated behaviour system that applies and extends existing movement-centric design principles to non-anthropomorphic robots. Our system combines a gaze-like attention engine with an arousal-modulated motion layer to explore how expressive and interactive behaviours influence social perception in robotic manipulators. In a user study, we find that robots exhibiting high attention—actively directing their focus toward users—are perceived as warmer and more competent, intentional, and lifelike. In contrast, high arousal—characterized by fast, expansive, and energetic motions—increases perceptions of discomfort and disturbance. Importantly, a combination of focused attention and moderate arousal yields the highest ratings of trust and sociability, while excessive arousal diminishes social engagement. These findings offer design insights for endowing non-humanoid robots with expressive, intuitive behaviours that support more natural human-robot interaction.
System Design
Our behaviour generation system integrates attention and arousal mechanisms to support expressive and socially responsive interaction in robotic manipulators. The system consists of two main components:
System architecture diagram for our proposed interactive and expressive robot behaviour system.
Attention Engine
The attention engine directs the robot's attention to salient features in its visual environment—specifically, users and individuals within range of an RGB-D sensor (Intel RealSense D435 Depth Camera). It computes an attention score for each detected individual, determining saliency and guiding the robot's gaze to enable dynamic, responsive interactions.
Each attention score is a weighted combination of positional cues, movement dynamics, and a habituation factor that prevents prolonged fixation on a single target. The score Φ is defined as:
Φ = wpP + wvV + Θ(t)
where P combines proximity and hand position factors, V captures torso and hand motion, and Θ(t) is a habituation factor that dynamically adjusts based on gaze history to maintain naturalistic attention shifts.
Robotic Gaze Algorithm
The gaze algorithm maps 3D target positions to joint-space postures, enabling the robot to orient toward salient individuals. Arousal modulates posture and motion dynamics: low arousal produces subdued, minimal motion, while high arousal increases speed, amplitude, and reach. A sinusoidal oscillation is applied to joint angles to simulate breathing, with oscillation intensity scaling with arousal to enhance realism and lifelikeness.
Attentional Drift Module
To avoid mechanical stiffness from continuous fixation, we introduce an Attentional Drift Module that injects naturalistic variability into the robot's gaze behaviour. The module generates transient virtual targets within the robot's workspace probabilistically, with likelihood increasing alongside arousal level, creating momentary attentional shifts without disrupting motion continuity.
Key Findings
We conducted a user study with 36 participants (18 males, 18 females, ages 18-25) examining how variations in arousal and attention influence human perceptions of a non-anthropomorphic robot. The study employed a 2 × 2 factorial design, crossing two levels of arousal (low and high) with two levels of attention (low and high).
Effects of Attention
High attention significantly enhanced perceptions across multiple dimensions:
- Warmth: Higher in high attention (M = 3.77) vs. low attention (M = 3.01), p < .001
- Competence: Higher in high attention (M = 5.11) vs. low attention (M = 3.06), p < .001
- Animacy: Higher in high attention (M = 4.10) vs. low attention (M = 3.44), p < .001
- Intentionality: Higher in high attention (M = 4.95) vs. low attention (M = 3.53), p < .001
- Sociability: Significantly improved with high attention, F(1, 35) = 7.22, p < .01
Effects of Arousal
High arousal had mixed effects:
- Discomfort: Increased in high arousal (M = 3.87) vs. low arousal (M = 3.34), p < .01
- Disturbance: Increased in high arousal, F(1, 35) = 7.58, p < .01
- Sociability: Decreased in high arousal, F(1, 35) = 4.75, p < .05
Interaction Effects
The combination of high attention and low arousal yielded the most positive social outcomes:
Interaction effect of arousal and attention on discomfort and sociability ratings.
- In high attention conditions, discomfort was significantly higher in high arousal (M = 3.71) than low arousal (M = 2.79), p < .01
- In high attention conditions, sociability was significantly lower in high arousal (M = 3.31) than low arousal (M = 4.17), p < .01
- Low arousal conditions resulted in longer interaction durations (M = 71.96s) than high arousal conditions (M = 62.21s), F = 4.66, p < .05
Key Insight: The most socially successful behaviours emerged when attention was high and arousal remained low. Users prefer systems that are attentive without being overstimulating, reflecting broader human norms around affect regulation where calm focus is associated with trustworthiness and emotional control.
Contributions
This work presents a unified framework for endowing non-anthropomorphic robots with expressive, intuitive behaviours. Our contributions include:
- Expressive Control System: We design an integrated system that combines arousal-based motion modulation with a real-time attention engine in a robotic manipulator, with natural motion continuity in mind.
- Empirical Evaluation: We conduct a factorial user study that evaluates how attention and arousal jointly shape perceptions of discomfort and sociability in non-anthropomorphic robots.
- Design Insights: We provide empirical insights into the social dynamics of expressive motion in non-humanoid robots, supporting their use in socially embedded contexts.
These findings offer concrete guidance for the design of socially intuitive robotic systems, particularly in non-humanoid forms, demonstrating that even robots without human-like features can convey character-like behaviour when designed with appropriate expressive strategies.
BibTeX
@article{el2025social,
title={The Social Life of Industrial Arms: How Arousal and Attention Shape Human-Robot Interaction},
author={El-Helou, Roy and Pan, Matthew KX},
journal={arXiv preprint arXiv:2504.01260},
year={2025}
}