AR Cue Reliability for Interrupted Task Resumption Affects Users' Resumption Strategies and Performance

Research Background and Problem

• What problems or challenges did the authors identify?

  • After task interruptions, people's efficiency and accuracy in resuming tasks often significantly decline, potentially leading to delays and errors.
  • While traditional environmental cues (e.g., objects in hand) can assist in task resumption, their effectiveness is limited. Augmented Reality (AR) cues offer greater flexibility but may not achieve 100% reliability due to technical limitations in practical applications.
  • The suboptimal reliability of AR cues (e.g., below 70%) could significantly impact task resumption performance, but few studies have explored how varying levels of reliability affect resumption strategies and outcomes.

• Why is this problem important?

  • Interruptions and task resumption are prevalent in high-risk domains such as healthcare, aviation, and manufacturing. Understanding and improving the performance of AR cues in these tasks is crucial for enhancing overall efficiency and reducing error risks.
  • Understanding how reliability levels influence the choice of resumption strategies and their relationship with task performance is key to designing effective resumption support tools.

• Research Motivation and Related Work

  • Existing studies have shown that 100% reliable AR cues are effective for task resumption, but achieving fully reliable AR systems in real-world scenarios is highly challenging.
  • Several studies have noted that the efficiency of automation and its enhancement of user task performance drop significantly when reliability falls below 70%.
  • This study systematically investigates, for the first time, the impact of AR cue reliability at different levels (64%, 86%, 100%) on task resumption through experimental methods.

Solution

• What methods or solutions did the authors propose?

  • Designed an experiment to study user performance in resuming interrupted physical sorting tasks, using AR cue reliability (100%, 86%, 64%, and no cues) as the primary variable.
  • Adopted the "Memory for Goals Theory" as the theoretical framework and quantified task resumption performance by measuring resumption lag and resumption errors.
  • Analyzed how AR cue reliability influences the resumption strategies chosen by users (e.g., relying on cues, recalling task progress).

• What is innovative about this solution?

  • For the first time, explored the impact of less-than-perfectly reliable AR cues on task resumption, building on prior research on 100% reliable cues.
  • Linked AR cue reliability to task resumption strategies, uncovering the dual influence of reliability on strategy selection and final task performance.
  • Refined comparative analysis across different reliability scenarios by separating conditions with "cue presence" and "cue absence."

• What are the implementation steps and key technologies used?

  1. Experimental Design: Based on a pill-sorting task (adapted from Bahnsen's research), participants were instructed to place pills of specific colors into designated boxes following AR cues.
  2. Interruption Task: Introduced 15-second or 45-second math tasks during the primary task to simulate real-world interruptions.
  3. Environment Setup: Used Microsoft HoloLens 2 to provide AR cues, displayed as red arrows indicating the pill to resume sorting.
  4. Variable Control: Set AR cue reliability levels (100%, 86%, 64%, or no cues) across different experimental groups.
  5. Data Analysis: Measured resumption lag and error rates, categorized participants' resumption strategies, and conducted statistical analysis to explore the relationship between reliability and strategies.

Research Findings

• What specific findings were obtained?

  • Task Resumption Performance: 100% reliable AR cues significantly reduced resumption lag and errors, especially after longer interruptions.
  • Impact of Suboptimal Reliability:
    • Overall, 86% and 64% reliable cues still outperformed the no-cue condition, but performance declined as reliability decreased.
    • Participants in the 86% reliability group relied more on "checking task status" (a less efficient strategy) during cue absence, without an increase in error rates.
    • Participants in the 64% reliability group relied more on "recalling task status" (faster but with higher error rates), performing worse than the no-cue group.
  • Strategy Selection: AR cue reliability significantly influenced participants' choice of resumption strategies, which in turn affected performance.

• What advantages does it have compared to existing solutions?

  • Explored task resumption performance and strategies under varying reliability levels, considering not only resumption time but also error rates and strategies.
  • Provided practical insights for designing AR cues with different reliability levels, helping to balance efficiency and accuracy.

• What were the experimental or evaluation results?

  • In the 100% reliability group, nearly all participants chose to "fully rely on cues," achieving the highest resumption efficiency and lowest error rates.
  • In the 64% reliability group, cue absence led to slower resumption and increased error rates, indicating that reliability below 70% may cause strategy and performance issues.
  • Even under 64% reliability conditions, task resumption performance during cue availability was significantly better than in the no-cue condition, demonstrating the partial effectiveness of suboptimal cues.

• Limitations and Future Directions

  • Limitations:
    • The experimental task was relatively simple and did not fully simulate the complexity of real-world work environments.
    • The frequency of interruptions was higher than in typical work scenarios, potentially limiting the natural development of some strategies.
    • The study used explicit "cue absence" failures rather than subtler "erroneous cues," which may have limited the impact on participants' trust and strategies.
  • Future Directions:
    • Validation studies in more complex scenarios (e.g., manufacturing processes or healthcare tasks).
    • Exploration of how different types of automation failures affect task resumption performance and user trust.
    • Integration of computational models to further predict task resumption performance under varying reliability conditions.

Conclusion

This study systematically explored the role of AR cues with varying reliability levels in task resumption after interruptions, revealing the profound impact of reliability on task resumption efficiency, error rates, and strategy selection. The findings not only confirm the partial effectiveness of suboptimal reliability cues but also provide critical guidance for optimizing AR cue design in real-world applications.