Large-Scale Multi-Robot Coverage Path Planning via Local Search

TL;DR

The paper tackles large-scale multi-robot coverage on a 2D grid by shifting from tree-cover STC methods on $G$ to a direct search on the decomposed graph $D$. It introduces Extended-STC (ESTC) to guarantee complete coverage on incomplete terrains and develops LS-MCPP, a local-search framework guided by boundary-editing operators that adjust subgraphs $D_i$. ESTC provides theoretically grounded coverage guarantees, while LS-MCPP uses operator pools, heuristics, and simulated-annealing-style acceptance to efficiently navigate the solution space, achieving substantial makespan reductions and orders-of-magnitude faster runtimes than tree-cover baselines. The approach demonstrates strong practical impact for large-scale real-world coverage tasks, particularly in environments with incomplete or partially known terrain data.

Abstract

We study graph-based Multi-Robot Coverage Path Planning (MCPP) that aims to compute coverage paths for multiple robots to cover all vertices of a given 2D grid terrain graph $G$. Existing graph-based MCPP algorithms first compute a tree cover on $G$ -- a forest of multiple trees that cover all vertices -- and then employ the Spanning Tree Coverage (STC) paradigm to generate coverage paths on the decomposed graph $D$ of the terrain graph $G$ by circumnavigating the edges of the computed trees, aiming to optimize the makespan (i.e., the maximum coverage path cost among all robots). In this paper, we take a different approach by exploring how to systematically search for good coverage paths directly on $D$. We introduce a new algorithmic framework, called LS-MCPP, which leverages a local search to operate directly on $D$. We propose a novel standalone paradigm, Extended-STC (ESTC), that extends STC to achieve complete coverage for MCPP on any decomposed graphs, even those resulting from incomplete terrain graphs. Furthermore, we demonstrate how to integrate ESTC with three novel types of neighborhood operators into our framework to effectively guide its search process. Our extensive experiments demonstrate the effectiveness of LS-MCPP, consistently improving the initial solution returned by two state-of-the-art baseline algorithms that compute suboptimal tree covers on $G$, with a notable reduction in makespan by up to 35.7\% and 30.3\%, respectively. Moreover, LS-MCPP consistently matches or surpasses the results of optimal tree cover computation, achieving these outcomes with orders of magnitude faster runtime, thereby showcasing its significant benefits for large-scale real-world coverage tasks.

Figures (10)

• Figure 1: Graph-based CPP and MCPP: Gray squares, black circles, and black stars represent terrain graph vertices, decomposed graph vertices, and initial vertices of robots, respectively; Solid lines and dashed lines represent coverage paths and spanning edges, respectively. (a) Terrain graph with uniform edge weights. (b) The single-robot coverage path generated by STC. (c)(d) Suboptimal and optimal $2$-robot coverage paths with makespans $2$ and $1.5$, respectively.
• Figure 2: A CPP instance with a root vertex (black star) and incomplete terrain vertices (yellow cells), where STC has no complete coverage. Full-STC is sub-optimal with undesirable routes (red circled area), and ESTC is optimal based on the augmented terrain graph $G'$.
• Figure 3: An example of grow operator $o_g(i,e)$ and the coverage paths after it is executed. Blue circles and black circles represent vertices of the boundary vertex set $B_i$ and vertices of $V_{d,i}$, respectively. (a) An invalid grow operator without any parallel edge in $D_i$. (b)(c) Two valid grow operators.
• Figure 4: An example of deduplicate operator $o_d(i,e)$. Red circles and black circles represent vertices in $V^+$ and not in $V^+$, respectively. (a) Four neighbors of $\delta_u$. (b) An invalid operator violating condition (3). (c)(d) Two valid operators.
• Figure 5: forcedDeduplication() executed on a coverage path $\pi_i$. Each red circle represents a duplication of $V_{d,i}$. In each frame, a U-turn $(u, v)\in\pi_i$ is found and $u, v$ are removed from $\pi_i$ until no U-turn exists.

Theorems & Definitions (1)

• theorem 1