A Combinatorial Bound for Beacon-based Routing in Orthogonal Polygons

Beacon attraction is a movement system whereby a robot (modeled as a point in 2D) moves in a free space so as to always locally minimize its Euclidean distance to an activated beacon (which is also a point). This results in the robot moving directly towards the beacon when it can, and otherwise sliding along the edge of an obstacle. When a robot can reach the activated beacon by this method, we say that the beacon attracts the robot. A beacon routing from $p$ to $q$ is a sequence $b_1, b_2,$ …, $b_{k}$ of beacons such that activating the beacons in order will attract a robot from $p$ to $b_1$ to $b_2$ … to $b_{k}$ to $q$, where $q$ is considered to be a beacon. A routing set of beacons is a set $B$ of beacons such that any two points $p, q$ in the free space have a beacon routing with the intermediate beacons $b_1, b_2,$ …, $b_{k}$ all chosen from $B$. Here we address the question of “how large must such a $B$ be?” in orthogonal polygons, and show that the answer is “sometimes as large as $[(n-4)/3]$, but never larger.”

💡 Research Summary

The paper investigates the problem of determining how many beacons are necessary to guarantee that any two points inside an orthogonal polygon can be connected by a beacon‑based routing sequence. A beacon‑based routing sequence consists of a series of beacons (b_1, b_2, \dots, b_k) that are activated one after another; each activation pulls the robot (modeled as a point) toward the active beacon by locally minimizing Euclidean distance, moving straight when possible and sliding along obstacle edges otherwise. If the robot reaches the beacon, we say the beacon attracts the robot. The relation “attracts” is asymmetric, but it contains the usual visibility relation and, in orthogonal polygons, it is a subset of staircase visibility (paths that are both (x)- and (y)-monotone).

The authors first review prior work: Biro et al. proved that for arbitrary simple polygons a minimum routing beacon set can be as large as (\lfloor (n-2)/2 \rfloor) and that finding the smallest such set is NP‑hard. They also conjectured that for orthogonal polygons the bound would be (\lfloor (n-4)/4 \rfloor). This paper disproves that conjecture and establishes the exact extremal bound (\lfloor (n-4)/3 \rfloor).

To obtain the bound, the polygon is placed in general position (no two edges are collinear) and a vertical decomposition (also called a trapezoidation) is performed by dropping a vertical chord from every reflex vertex. This partitions the polygon into ((n-2)/2) axis‑aligned rectangles, whose adjacency forms a tree (the dual graph). Each rectangle’s neighbors are classified by side (left/right, top/bottom), by height (tall vs. short), and by whether a side has a single neighbor (solo) or two (paired).

The key technical notion is that of a local routing segment: a maximal portion of a routing path that lies within at most three rectangles of the decomposition. A routing path is local if all its segments are local, and a beacon set is local if it enables a local routing path between any pair of points. The authors show that a local beacon set can be constructed by placing beacons so that each rectangle (and its neighboring rectangles) is “covered” in a symmetric sense: a beacon (b) covers a region (R) if every point in (R) both attracts and is attracted by (b).

Four lemmas describe how to achieve coverage with a small number of beacons:

1. Lemma 1 handles a solo short neighbor (S) of a rectangle (R); any point inside (R) locally covers (S) (except possibly a reflex “curl” vertex), and vice‑versa.
2. Lemma 2 extends this to a solo neighbor when a beacon lies inside the rectangle containing (R) and the beacon’s rectangular hull stays inside the polygon.
3. Lemma 3 treats tall solo neighbors under a similar hull condition, provided the reflex edges of the combined L‑shaped region lie on the polygon boundary.
4. Lemma 4 deals with paired neighbors on the same side; a beacon placed in the “modified center” of the rectangle (the full‑width strip minus the two corners adjacent to the paired neighbors) simultaneously covers both paired neighbors.

These lemmas guarantee that, for any rectangle and its immediate neighbors, a constant (at most three) number of beacons suffices to achieve mutual coverage, thereby eliminating “trapped paths” where a robot would become stuck on a boundary edge.

The upper bound proof proceeds by induction on the dual tree. Removing a leaf rectangle reduces the number of vertices by two; the induction hypothesis supplies a beacon set for the smaller tree with at most (\lfloor (n’ -4)/3 \rfloor) beacons, where (n’) is the number of vertices of the reduced polygon. By applying the appropriate lemma to the removed leaf and possibly adding a single new beacon, the authors extend the set to the original polygon while preserving the bound. Consequently, any orthogonal polygon with (n) vertices admits a routing beacon set of size at most (\lfloor (n-4)/3 \rfloor).

To show that this bound is tight, the paper presents a constructive family of orthogonal polygons. Starting from a base rectangle, the construction repeatedly attaches a “staircase” gadget consisting of three new vertices. Each gadget forces the addition of at least one new beacon in any routing set, because the gadget creates a region that cannot be covered by beacons placed elsewhere without violating the local coverage conditions. After adding (k) gadgets, the polygon has (n = 3k + 4) vertices and requires exactly (k = \lfloor (n-4)/3 \rfloor) beacons.

The results settle the extremal question for orthogonal polygons: the minimum size of a universal routing beacon set is bounded above by (\lfloor (n-4)/3 \rfloor) and this bound is achievable. The paper also contributes methodological tools—local routing segments, the classification of rectangle neighbors, and the coverage lemmas—that may be useful for related problems such as beacon‑based guarding, attraction kernels, and greedy geographic routing in sensor networks. Future work could explore extensions to non‑orthogonal polygons, three‑dimensional polyhedral environments, or dynamic beacon activation strategies.