Clustering to Determine Interconnected Activities in Supervisory Control Tasks of Pilots

Abstract

We propose a method that allows pilot activity determination. Such systems are of great interest for assistance systems that adapt to pilot performance. However, the determination of supervisory control tasks is non-trivial since they can only be observed indirectly through pilot actions. Our previous work to determine activities based on evidential reasoning resulted in a highly fragmented pattern of recognized tasks over time. To address this, we suggest clustering these scattered patterns into partitions which better reflect the activities of the pilot. To evaluate the proposed approach, we conducted an experiment in a fast-jet simulator with 11 participants. Using our former approach, we attempted to determine the activities and then applied k-Means clustering to find partitions of interconnected activities. Lastly, we evaluated whether the results could be compared to the activities as reported by the participants. Our results indicate that clustering may not be an effective activity determination method for adaptive assistance systems in Manned-Unmanned Teaming applications.

Keywords: Activity Recognition, Human Factors, Pattern Recognition, Clustering, Unsupervised Learning, Adaptive Assistance

DOI: 10.54941/ahfe1004329