Human-Robot Collaboration: The Trade-off Between Efficiency and Trust

A quantitative experimental study (N=58) evaluating how automation levels impact operator trust and cognitive workload in industrial settings.

Role: Principal Investigator (Thesis Research)

Timeline: Aug 2023 – Dec 2024

Team: Faculty Advisor (Dr. Laura Ikuma), Thesis Committee (Dr. Isabelina Nahmens, Dr. Corina Barbalata)

Tools & Methods: Universal Robots (UR5e), PolyScopeX, Python (Data Scripting), R (Statistical Analysis), True Experimental Design, A/B Testing, Physiological Monitoring (Heart Rate), NASA-TLX, Trust in Automation (TiA) Scale

Executive Summary:

Challenge: Industry 4.0 is introducing “Cobots” (Collaborative Robots) to shared workspaces to reduce physical strain on workers. However, a lack of human factors consideration can lead to low adoption rates and operator frustration. We needed to understand: Does higher automation actually lead to a better user experience?

Approach: I designed a controlled laboratory experiment with 58 participants to compare two interaction modes: Semi-Automated (Human-in-the-loop) vs. Fully Automated (Machine-paced). I utilized physiological sensors (HR monitor, ) and standardized psychometric scales (Trust in Automation (TiA) questionnaire, NASA Task Load Index Scale) to measure the invisible “cognitive costs” of collaboration.

Outcome: The data revealed a critical trade-off: While Fully Automated modes were 22.3% faster , they caused cognitive disengagement and lower trust. Semi-Automated modes, while slower, resulted in significantly higher user trust and better cognitive engagement. The study recommends balancing automation with user control to optimize performance.

Context: From “Pain” to “Protocol”

Inspiration: This research was inspired by real-world observation. While visiting Setex Inc. (an automotive manufacturer), I observed a car seat upholsterer using a pneumatic gun repetitively, which she noted caused chronic wrist pain. While robots could help, there was a clear separation between human tasks and robot tasks, sparking my interest in how to merge them effectively.

Research Question: How do we design robot behaviors that reduce physical load without increasing mental load or eroding trust?

Experimental Design (Mixed-Methods)

Task: Participants collaborated with a Universal UR5e Robotic Arm to assemble a miniature lamp post using color-coded blocks.

The Variables:

Group A (Semi-Automated): The user controlled the pace. The robot waited for a signal from the user to initiate its task.
Group B (Fully Automated): The robot moved on a fixed timer, forcing the user to adapt to the machine’s pace.

The Metrics (Triangulated Data):

Physiological: Heart Rate Variability (HRV) to measure real-time stress.
Performance: Detection Response Time (DRT) to measure cognitive “spare capacity” (attention).
Subjective: NASA-TLX (Workload) and Trust in Automation (TiA) surveys.

Key Insights & Data Analysis

Used R to perform T-tests and correlation analysis, I uncovered three major findings:

The Efficiency vs. Trust Paradox

Automation isn’t always “better.”

Data: The Fully Automated group was 22.3% faster in assembly time (p<0.0001).
Catch: However, the Trust in Automation subscale showed significantly higher ratings in the Semi-Automated group (p=0.0067).
UX Takeaway: Speed is a business metric; Trust is a user metric. Forcing speed without control kills trust.

The “Cognitive Disengagement” Danger

Data: In the Automated group, we found a positive correlation between Heart Rate and Detection Response Time (r =0.29, p =0.05), meaning higher arousal led to slower reaction times.
Insight: When users had no control (Fully Auto), they experienced “cognitive disengagement.”²³. Because the robot moved on its own, users “zoned out” or became overloaded, making them less responsive to auditory alerts.
UX Takeaway: Keeping the human “in the loop” (Semi-Auto) maintains situational awareness²⁴.

Control Drivers Satisfaction

Data: Overall Trust ratings were 4.6% higher in the Semi-Automated mode.
The Insight: Users prefer agency. Being able to signal the robot (the “Go” button) made the system feel like a partner, whereas the automated timer made it feel like a taskmaster.

Design Recommendations

Based on this data, I developed specific guidelines for Human-Robot Interface (HRI) design:

Prioritize Agency over Speed: In complex or high-risk environments, sacrificing pure speed for user control (Semi-Automation) fosters higher trust and better long-term engagement
Implement Adaptive Control: Systems should not be binary (Manual vs. Auto). They should offer adaptive control options allowing users to adjust automation levels based on task complexity.
Transparent Feedback Loops: Fully automated systems need transparent feedback to prevent the “disengagement” we observed, ensuring users remain aware of the system’s intent.

Impact:

Academic Contribution: Published a comprehensive thesis contributing to the body of knowledge on Ergonomics and Industrial Engineering.
Methodological Framework: Validated a dual-task methodology (Assembly + DRT) for measuring cognitive load in physical workspaces, a novel approach for this specific type of collaborative task.
Future Application: These findings provide a data-backed foundation for designing safer “Cobot” workflows in manufacturing and healthcare sectors.