Robots are good at performing repetitive tasks with precision but lack intelligence. On the other hand, humans are better at handling unexpected situations, identifying abstractions, and possessing complex cognitive processing skills (Vysocky & Novak, 2016). Combining these complementary skills can result in a team capable of performing complex tasks with high combined system efficiency. However, robots are prone to unexpected failures impacting human-robot collaboration (HRC). Trust is one of the foundational factors in HRC, and it can impact system performance, acceptance, and utilization (Hopko, Wang, & Mehta, 2022). In order to achieve optimal operations, it is necessary to calibrate trust between the robot and the operator (de Visser et al., 2020). Furthermore, operators in an industrial setting are often subject to extended and erratic shifts, leading to both cognitive and physical fatigue even in automated environments. Understanding the interaction of cognitive fatigue and trust can help improve the HRC. Sixteen participants were recruited for the study after approval by the local IRB (IRB2020-0097DCR). The experiment used a UR10 (Universal Robots, DK) robot whose end-effector was controlled by a joystick input. Participants completed an S-shaped metal surface polishing task using the joystick. The experiment was conducted in two sessions (fatigue, no fatigue). Participants underwent two levels of robot reliability at each session (reliable and unreliable). Reliable trials were always presented first to build trust, and unreliable trials were presented to breach trust. During the unreliable trials, the robot was intentionally perturbed, and a reliability rate of 76% was used. Every participant completed 40 experimental trials in total, with 10 trials in each condition. To manipulate fatigue, participants underwent an hour-long computer-based sustained 2-back task before starting the polishing task, which has been shown to simulate cognitive fatigue (Hopstaken, van der Linden, Bakker, & Kompier, 2015). Throughout, we monitored subjective responses (trust, fatigue, NASA TLX), performance (efficiency, accuracy, precision), and brain activity using functional nearinfrared spectroscopy (fNIRS). Participants in fatigue conditions reported higher fatigue (p=0.003, η2=0.129) compared to no fatigue conditions. Higher fatigue (p<0.001, η2=0.067) was also reported in the unreliable conditions (4.474±2.464). Reliability had a significant effect on TRUST (p=0.005, η2=0.040); participants rated higher TRUST in reliable conditions. Reliability significantly impacted the activations in the LBA (p=0.03, η2=0.136), LDLPFC (p=0.041, η2=0.048), MDLPFC (p=0.03, η2=0.145), and RBA (p=0.036, η2=0.035), with all regions exhibiting higher activation in unreliable conditions. Fatigue significantly impacted the MDLPFC (0.016, η2=0.073) with higher activations in no fatigue conditions. M1 region showed a significant effect of fatigue×reliability (p=0.048, η2=0.017), no posthoc comparisons were significant (all p’s>0.172). RBA also had significant reliability×sex interaction (p=0.036, η2=0.035), post hoc tests revealed females in unreliable condition had higher activation than females in reliable condition (pTukey = 0.023). Further, fatigue×reliability×sex had a significant effect (p=0.026, η2=0.033), post-hoc test revealed females under fatigue had higher activation (pTukey=0.022) in unreliable conditions. Causal relationships between the regions were calculated using the MVGC toolbox (Barnett & Seth, 2014). Significant causal connections among the three brain regions increased from 1 in no-fatigue to 2 in fatigue under reliable conditions. However, connections decreased from 6 in no-fatigue to 4 in fatigue under unreliable conditions. Reliable conditions showed higher speed, lower deviation, and lower variance than the unreliable condition (all p’s<0.001). Increased activation in the LBA, LDLPFC, MDLPFC, RBA, and LBA regions of the brain corroborates earlier findings suggesting increased task difficulty and mental effort cause an increase in oxygenated hemoglobin level in the PFC (Causse, Chua, Peysakhovich, Del Campo, & Matton, 2017). Future robot designs with higher automation must consider these cognitive processes to design collaborations with reduced cognitive demands. Reduced activation in the MDLPFC region is linked to a decline in working memory, suggesting participants were distracted and less focused on the task. Increased activations in RBA have been attributed to the increase in visuospatial cognition, which helps identify the visual and spatial relationship between the objects. Insights from this investigation can advance objective trust measures under different states and help design better collaborations.
[1]
Ranjana K. Mehta,et al.
Human Factors Considerations and Metrics in Shared Space Human-Robot Collaboration: A Systematic Review
,
2022,
Frontiers in Robotics and AI.
[2]
Mark A. Neerincx,et al.
Towards a Theory of Longitudinal Trust Calibration in Human–Robot Teams
,
2019,
International Journal of Social Robotics.
[3]
Mickaël Causse,et al.
Mental workload and neural efficiency quantified in the prefrontal cortex using fNIRS
,
2017,
Scientific Reports.
[4]
Petr Novák,et al.
HUMAN – ROBOT COLLABORATION IN INDUSTRY
,
2016
.
[5]
A. Bakker,et al.
A multifaceted investigation of the link between mental fatigue and task disengagement.
,
2015,
Psychophysiology.
[6]
Anil K. Seth,et al.
The MVGC multivariate Granger causality toolbox: A new approach to Granger-causal inference
,
2014,
Journal of Neuroscience Methods.