Who is in Control? Understanding User Agency in AR-assisted Construction Assembly

Research Background and Problem

• Identified Issues or Challenges:
  The authors discovered that automated content adaptation in AR (Augmented Reality) systems can significantly improve user operational efficiency and task outcomes. However, such automation also alters users' control and task autonomy. This automation may lead to user confusion regarding system behavior and a perceived reduction in decision-making authority. This trade-off has not been adequately studied in AR-assisted systems.

• Why This Problem Matters:
  Investigating user control and autonomy in AR systems is crucial for enhancing user experience, improving task efficiency, and supporting prolonged work durations. This is particularly important in the construction industry, which faces labor shortages and high workplace injury rates. Optimizing human-machine collaboration and automated systems can help address these industry bottlenecks.

• Research Motivation and Related Work:
  Although researchers have extensively studied AR system performance and content adaptation, the impact on user agency has been rarely explored. Existing literature primarily focuses on overall system performance rather than the psychological, task efficiency, and job satisfaction effects of diminished user control. This study aims to explore the trade-offs between high agency (high user control) and low agency (low user decision-making involvement) in the complex context of construction assembly.

Solution

• Proposed Solution:
  The authors designed two interaction modes for a head-mounted AR-assisted system:

  1. High Agency Mode: Users actively trigger system behaviors, such as manually selecting task prompt steps.
  2. Low Agency Mode: The system automatically triggers task prompts at fixed intervals without user input.

  The study encompasses both cognitively intensive and repetitive physical tasks, evaluating user experience, task performance, psychological needs, and workload across the two modes.

• Innovative Contributions:

  1. Introducing research on user agency and control into construction task scenarios, a rarely studied domain.
  2. Innovatively combining psychological needs (autonomy, competence) with workload (NASA-TLX mental workload index) to assess the impact of AR design on users.
  3. Integrating experimental results with ecological validity through semi-structured interviews, combining laboratory findings with feedback from domain experts in real-world work scenarios for multi-perspective analysis.

• Implementation Steps and Techniques:

  1. Designing two construction assembly tasks: cognitively intensive (e.g., complex wooden joint assembly) and physically intensive (e.g., heavy object lifting and placement).
  2. Developing a head-mounted AR system using the Unity platform and Rhino 3D modeling tools to generate task content, with real-time interaction controlled via WebSocket.
  3. Recruiting 24 experimental participants (university students) and 8 domain experts (professional carpenters) to complete tasks under both agency modes.
  4. Collecting data through questionnaires (e.g., NASA-TLX, SUS, BPNSS) and semi-structured interviews.

Research Findings

• Specific Findings:

  1. Laboratory Study: The low agency mode reduced mental workload but decreased user autonomy and system usability ratings in cognitive tasks.
  2. Domain Expert Interviews: Compared to factory daily tasks, the study tasks involved fewer large-element operations, high variability, and real-world challenges in complex site configurations.

• Advantages Over Existing Solutions:
  By closely integrating user agency research with the construction industry, the study provides practical design guidelines for future AR systems tailored to target user needs. It also highlights the potential of low agency systems in training novice users.

• Experimental or Evaluation Results:

  1. Mental Workload: The low agency mode significantly reduced mental workload, especially in physical and complex cognitive tasks.
  2. Autonomy: The high agency system received higher autonomy ratings in more complex tasks.
  3. Performance: Task completion performance showed no significant differences between the two modes, but the high agency system demonstrated faster step completion times.
  4. User Feedback: Domain experts preferred prioritizing autonomous control, though the automation features of the low agency mode showed advantages for novice users and specific scenarios.

• Limitations and Future Directions:

  1. The sample size was relatively small, with experimental participants and domain experts primarily drawn from a single construction company and student group. Future research should expand to more diverse user groups and construction professions.
  2. Long-term effects of using low agency systems on skill acquisition and psychological perception were not fully assessed, suggesting the need for longitudinal studies.
  3. The complexity and environmental characteristics of work tasks were not comprehensively covered. Future studies should include more complex task types, such as large component operations and team collaboration.

This study provides critical insights into AR design for the construction industry, emphasizing the delicate balance between user agency and automated systems. Future research can further explore the impact of human-computer interaction in prolonged tasks while offering richer ecological validation for designs across diverse usage scenarios.