On approximating shortest paths in weighted triangular tessellations

TL;DR

This paper analyzes how well weighted shortest paths are approximated in discretized weighted triangular tessellations, providing tight bounds that are independent of weight assignments and demonstrating near-optimal approximation ratios.

Contribution

It establishes tight, weight-independent bounds on the approximation ratios of grid and vertex shortest paths in weighted triangular tessellations, extending previous work.

Findings

Worst-case ratio for grid path approximation is approximately 1.15.

The bounds are independent of face weight assignments.

The results are tight and extend prior grid-based path analysis.

Abstract

We study the quality of weighted shortest paths when a continuous 2-dimensional space is discretized by a weighted triangular tessellation. In order to evaluate how well the tessellation approximates the 2-dimensional space, we study three types of shortest paths: a weighted shortest path $ \mathit{SP_w}(s,t) $, which is a shortest path from $ s $ to $ t $ in the space; a weighted shortest vertex path $ \mathit{SVP_w}(s,t) $, which is an any-angle shortest path; and a weighted shortest grid path $ \mathit{SGP_w}(s,t) $, which is a shortest path whose edges are edges of the tessellation. Given any arbitrary weight assignment to the faces of a triangular tessellation, thus extending recent results by Bailey et al. [Path-length analysis for grid-based path planning. Artificial Intelligence, 301:103560, 2021], we prove upper and lower bounds on the ratios $ \frac{\lVert \mathit{SGP_w}(s,t)\rVert}{\lVert \mathit{SP_w}(s,t)\rVert} $, $ \frac{\lVert \mathit{SVP_w}(s,t)\rVert}{\lVert \mathit{SP_w}(s,t)\rVert} $, $ \frac{\lVert \mathit{SGP_w}(s,t)\rVert}{\lVert \mathit{SVP_w}(s,t)\rVert} $, which provide estimates on the quality of the approximation. It turns out, surprisingly, that our worst-case bounds are independent of any weight assignment. Our main result is that $ \frac{\lVert \mathit{SGP_w}(s,t)\rVert}{\lVert \mathit{SP_w}(s,t)\rVert} = \frac{2}{\sqrt{3}} \approx 1.15 $ in the worst case, and this is tight. As a corollary, for the weighted any-angle path $ \mathit{SVP_w}(s,t) $ we obtain the approximation result $ \frac{\lVert \mathit{SVP_w}(s,t)\rVert}{\lVert \mathit{SP_w}(s,t)\rVert} \lessapprox 1.15 $.