Exploring Human-Robot Collaboration: Analysis of Interaction Modalities in Challenging Tasks

Technical Summary: Exploring Human-Robot Collaboration: Analysis of Interaction Modalities in Challenging Tasks

Problem Statement
As robots increasingly enter social environments such as healthcare, logistics, and education, the challenge shifts from mere task execution to socially acceptable collaboration. A critical gap exists in understanding how different levels of robot proactivity—specifically reactive (request-based) versus proactive (initiative-based)—impact user perception, emotional state, and task performance. While existing literature suggests that proactive behavior can enhance fluency and anthropomorphic perception, other studies indicate users may find such interventions disturbing or intrusive, preferring to retain control over intervention timing. This study investigates these conflicting dynamics to determine which interaction modality users prefer and how proactive versus reactive behaviors influence the human experience in a cognitively and physically demanding collaborative task.

Methodology
The study employed a within-subjects experimental design involving 18 adult participants. The core task required participants to assemble a seven-layer colored tower from memory using LEGO DUPLO bricks. The task was designed to be cognitively demanding (requiring memorization of a color sequence shown briefly) and physically challenging (requiring participants to stand and retrieve specific bricks from the opposite side of the table).

Participants completed the task under three distinct modalities:

1. Passive (PA): The participant worked alone with no robotic assistance.
2. Reactive (R): A mobile robot (DJI RoboMaster EP Core) provided assistance only upon explicit verbal request. Assistance included fetching specific bricks or suggesting the correct color via LED cues and speech.
3. Proactive (PR): The robot autonomously intervened without explicit requests. It fetched the fourth brick of a new layer immediately after the first was placed and signaled errors (e.g., incorrect color) via red LED flashes and verbal alerts. Users could still request color suggestions.

To ensure smooth execution while isolating user responses to assistance timing from perception robustness, a Wizard-of-Oz setup was utilized, where an operator discreetly teleoperated the robot. Data collection included objective performance metrics (completion time, error counts, request frequency) and subjective metrics gathered via a 25-item questionnaire based on a 7-point Likert scale. The questionnaire assessed task perception (difficulty, mental/physical demand), emotional experience (anxiety, comfort, control), and robot perception (trust, usefulness, reliability, interaction quality).

Key Contributions

• Comparative Analysis of Modalities: The study provides a direct comparison of passive, reactive, and proactive modalities within a single experimental framework, extending previous work (e.g., Baraglia et al., 2017) by introducing a passive baseline and increasing task complexity through memory and physical retrieval requirements.
• Holistic Evaluation Framework: The research integrates objective performance data with a granular subjective assessment, grouping metrics into "task," "emotional," and "robot" scores to evaluate the multidimensional impact of robot behavior.
• Wizard-of-Oz Implementation for HRI: The study demonstrates a controlled teleoperation setup that mimics autonomous behavior to test user reactions to proactive support without the technical limitations of current autonomous perception systems.

Results

• Performance Metrics: Contrary to the hypothesis that assistance would speed up the task, the Passive modality resulted in the fastest completion times. Both Reactive and Proactive modalities increased completion time, attributed to interaction overhead (waiting for robot motion and feedback). However, the difference in completion time between Reactive and Proactive modalities was not statistically significant.
• Subjective Task Perception: Participants rated the task as significantly easier and less physically demanding in the Proactive modality compared to the Passive condition. The Proactive modality also showed trends toward higher accuracy and lower anxiety compared to the Passive condition.
• Robot Perception and Preference:
  • Preference: 67% of participants preferred the Proactive modality, and 78% judged it as the most useful.
  • Trust and Usefulness: The Proactive modality received significantly higher scores for robot usefulness, reliability, and trust compared to the Reactive modality.
  • Control: Despite the preference for proactive support, a subset of participants (28%) preferred the Reactive modality, citing a desire for greater control over the robot's interventions.
• Emotional Experience: While emotional scores showed only marginal statistical improvements in the Proactive condition compared to Passive, the overall trend indicated a more positive affective state during assisted interactions.

Significance and Claims
The paper claims that timely proactive support can significantly improve the user experience in controlled collaborative tasks, even when it does not improve objective efficiency. The results suggest that users are willing to trade task completion speed for reduced physical effort and enhanced interaction quality. The study concludes that proactive robots are generally perceived as helpful and non-intrusive in this context, challenging the notion that users universally prefer to maintain strict control over intervention timing.

However, the authors maintain a modest stance, acknowledging the small sample size (N=18) and the specific constraints of the experimental setup. They posit that while proactive behavior is beneficial for user experience in this controlled environment, future research is required to validate these findings in more complex, dynamic scenarios and across diverse user populations with varying levels of prior robot experience. The work supports the design of assistive robots that engage proactively while maintaining transparency and user agency, but emphasizes the need for adaptive modalities to accommodate individual differences in collaboration styles.