The Role of Task Frequency and Complexity in Remote Assistance for Highly Automated Vehicles: Assessing Mental Load based on Eyetracking and Physiology

Abstract

Guidance through remote assistants (RAs) is a key component for introducing driverless highly automated vehicle fleets (SAE level 4) into future mobility systems. RAs likely encounter situations with variable task demands depending on the frequency of incoming tasks and the complexity of the problems they encounter, potentially inducing mental underload or overload that affect performance during operation. To analyze this, we conducted a simulator study where 19 participants (5 f, 13 m, 1 d) took over the role of RAs for an urban automated shuttle to determine how sensor-based workload prediction can be utilized to differentiate task frequency and complexity variations during remote assistance. Using a gradient boosting classifier, we find that that multi-class multi-output taskload classification using physiology- and eyetracking-based indicators achieves strong results across participants (ROC-AUC - MN: 0.90, SD: 0.07), forming a basis for providing situation-specific adaptive support during remote operation.