Measuring Trust for Exoskeleton Systems

Leia Stirling, Man I Wu, Xiangyu PengarXiv 2024 · DOI · arXiv

Abstract

Measuring trust in wearable robots, such as exoskeletons, requires a specialized approach due to the "tight coupling" between human and machine movements. Traditional trust models designed for non-wearable autonomous systems often fail to capture the physical, bidirectional flow of force inherent in wearable technology. This paper proposes a conceptual framework that moves beyond general surveys, offering specific perceptual and behavioral metrics—including muscle activation and gaze tracking—to assist engineers in designing more reliable and intuitive wearable systems.

Introduction

Most research on human-robot trust focuses on "non-wearable" systems like autonomous vehicles or industrial arms. In these scenarios, humans observe and react from a distance. Exoskeletons present a unique challenge: because they are worn, timing errors as small as 2.8% of a walking stride can be immediately felt. A lack of trust leads users to actively fight the device's motors, resulting in increased physical effort, potential injury, or abandonment of the technology. There is a clear need for actionable metrics that capture this unique physical trust relationship.

Methodology: A Framework for Operationalizing Trust

The proposed framework adapts the classic "Purpose, Process, and Performance" model of trust to exoskeleton-specific attributes. This approach emphasizes two parallel paths for measurement: user perceptions and direct behavioral surrogates.

Perceptual Metrics

Refining survey methodology is crucial for obtaining actionable engineering data. Instead of broad questions about general trust, the framework suggests targeted inquiries into specific system behaviors. For instance, users might rate how effectively the exoskeleton aligns with their expectations or supports their motor actions.

Table 1: Perceptual attributes of trust in exoskeleton systems, mapping dimensions to specific survey-based attributes.

Behavioral and Direct Measures

Physical signals often reveal trust levels that users may not consciously perceive. For example, "muscle co-contraction" (tensing up) or altered joint kinematics (such as increased peak hip flexion) can indicate a user's attempt to compensate for a robot's perceived unpredictability. Gaze tracking also serves as a proxy for understanding; frequent monitoring of the device often correlates with confusion or a lack of trust in the system's operation.

Table 2: Direct behavioral measures of trust, connecting trust dimensions to observable data like muscle activation and gaze tracking.

Key Insights

  • The Illusion of Relaxation: Studies show users may believe they are relaxed while EMG data reveals active muscle resistance, highlighting a misalignment between perception and control.
  • Actionable Specificity: By breaking trust into sub-dimensions like Integrity or Dependability, designers can pinpoint technical issues such as sensor latency or mechanical support.
  • Dynamic Nature of Trust: Trust is not static; it fluctuates based on task complexity and environmental factors, such as moving from flat ground to uneven terrain.

Results and Discussion

The primary contribution of this work is the categorization of trust attributes into comprehensive tables for perceptions and direct measures. Experimental validation has shown that as controller error rates increase, user ratings of "predictability" and "supportiveness" decrease, confirming that these metrics are sensitive to actual system performance.

However, challenges remain. These metrics require further validation to distinguish between trust and general physical discomfort or fatigue. Additionally, individual differences in risk tolerance and prior experience with technology must be better integrated into the framework to provide a holistic view of human-robot interaction.

Conclusion

Moving toward a multi-dimensional measurement of trust—combining subjective surveys with objective physiological data—provides engineers with the actionable outcomes necessary to improve exoskeleton design. Future work will focus on refining these metrics to account for individual human variability and long-term trust dynamics in real-world environments.

Summary updated

BibTeX
@misc{stirling2024measuring,
  author = {Leia Stirling and Man I Wu and Xiangyu Peng},
  title = {{Measuring Trust for Exoskeleton Systems}},
  year = {2024},
  eprint = {2407.07200},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2407.07200},
}
Figure