Trust in Autonomous Human–Robot Collaboration: Effects of Responsive Interaction Policies

Authors: Shauna Heron (Laurentian University), Meng Cheng Lau (Laurentian University)

Abstract

This pilot study examined how interaction policies influence trust during in-person collaboration with a fully autonomous social robot. Participants engaged in dialogue-driven tasks with either a responsive robot (adapting based on inferred interaction state) or a neutral, reactive robot. Responsive interaction was associated with significantly higher post-interaction trust under viable communication conditions, despite no reliable differences in overall task accuracy. The study highlights that trust emerges from interaction process dynamics rather than task performance alone, and operates within communication viability constraints.

Key Findings

Trust Outcomes:

• Responsive condition showed substantially higher trust scores on both measures
• Trust in Industrial Human–Robot Collaboration scale (TI-HRC) differed by ~26 points
• Trust Perception Scale–HRI (TPS-HRI) differed by ~15 points
• Effects held regardless of overall task accuracy

Interaction Quality:

• Robot employed encouragement in 36% of responsive dialogue versus 0% in control (p < .001)
• Responsive robot expressed empathy in 13% of turns versus 0% in control (p < .001)
• Proportions of communication breakdown did not differ between conditions (25% vs. 22%)

Communication Viability: Five sessions (out of 29) exhibited severe communication failure and were excluded from primary analyses. Trust advantages for responsive interaction attenuated significantly when speech recognition failures accumulated.

Methods

Design: Between-subjects experiment comparing two interaction policies in a fully autonomous system

Participants: 29 community members (24 in eligible sample after communication failure exclusions)

Robot Platform: Misty-II mobile robot with integrated dialogue management, affect inference, and autonomous task progression

Task Structure:

• Task 1: Robot-dependent collaborative reasoning (suspect identification puzzle)
• Task 2: Open-ended problem-solving (location identification from technical logs)

Measures:

• Post-interaction trust assessments using TPS-HRI and TI-HRC scales
• Objective metrics: dialogue turns, response latency, engagement detection
• Dialogue coding for communication breakdowns and interaction quality

Main Conclusions

The study provides evidence that affect-responsive interaction policies can influence trust during fully autonomous, in-person human–robot collaboration. However, this effect operates within important boundaries. Communication viability emerged as a prerequisite rather than merely a moderator—when linguistic grounding collapsed, policy differences diminished.

Trust advantages were most pronounced under viable interaction conditions and attenuated when communication breakdown accumulated. The findings reinforce that trust reflects interaction quality dynamics rather than instrumental success, suggesting future systems should prioritize stable language handling and adaptive communication management alongside affective responsiveness.

Limitations

• Pilot sample size limited statistical power
• Substantial variability in functional spoken-language proficiency among participants
• Lack of non-embodied comparison condition
• Affect inference relied primarily on speech and conversational context

Future Directions

Authors recommend larger confirmatory studies, improved dialogue management architectures with explicit repair strategies, adaptive language complexity mechanisms, and evaluation under conditions exposing rather than concealing system limitations.