From Trajectories to Relations: Interpreting Overtake Manoeuvres with Large Language Models and Qualitative Spatial Representations

Supplementary materials for the short paper “From Trajectories to Relations: Interpreting Overtake Manoeuvres with Large Language Models and Qualitative Spatial Representations”. This archive contains the derived materials used to support an exploratory study of LLM-based interpretation of qualitative spatial encodings of an overtaking manoeuvre. The study represents one selected simulated traffic configuration, configuration 52, at three levels of representational complexity: 2, 5, and 10 moving points. For each representation level, the archive includes qualitative PDP-based inequality matrices for the longitudinal and lateral descriptors, heatmap visualisations, static trajectory plots, animations, structured API request and response files, and merged LLM interpretation outputs. The archive also includes the Python script used to construct descriptor-specific API payloads from labelled matrix CSV files and to run the staged interpretation workflow. In this workflow, each qualitative inequality matrix is converted into ordered same-time POI-pair summaries, which are submitted to the LLM separately for the longitudinal and lateral descriptors. The resulting partial interpretations are then merged into a final manoeuvre interpretation. Expert evaluation templates and completed expert assessments are also included. The raw simulated trajectory data are not included because of source restrictions. The shared files contain only derived qualitative encodings, visualisations, API materials, code, and evaluation documents needed to inspect and reproduce the interpretation workflow reported in the paper.