Assess sensitivity to the number of scenario positions

Determine how changing the number of discrete ego-centered grid positions N affects the validity, criticality, diversity, and collision-finding performance of the Teach-to-Crash scenario representation.

Background

The position-based scenario representation uses an ego-centered grid with N discrete position identifiers; the evaluation fixes N=9. Although the grid resolution directly controls placement granularity, alternative values of N are not experimentally evaluated.

A sensitivity analysis would establish whether the reported results depend on the selected spatial discretization and whether finer or coarser grids improve scenario generation.

References

We choose $K{=}6$ steps to divide the rollout into six 10\,s segments, sufficient for an approach, interaction, and resolution phase; we did not sweep alternative $K$, leaving this for future work (Section~\ref{sec:discussion}).

— Teach-to-Crash: A Closed-Loop Student-Teacher LLM Framework for Collision-Inducing Test Scenario Generation  (2609.27296 - Ghazal et al., 23 Sep 2026) in Section 3.4, Scenario Representation