Structural Induced Exploration for Balanced and Scalable Multi-Robot Path Planning

TL;DR

This work tackles the challenge of scalable and fair multi-robot path planning by embedding a persistent structural backbone into Ant Colony Optimization. The method, SINE, constructs a minimum-cost spanning skeleton as a backbone, biases pheromone updates toward backbone edges, and decomposes tasks for parallel solver execution to achieve balanced workloads and shorter collective routes. Empirical results on TSPLIB benchmarks and an Antarctica dataset show that SINE consistently outperforms standard ACO variants in total path length and workload balance, with strong scalability across problem sizes and robot counts. The framework also provides theoretical and practical insights into convergence behavior and interpretable structure-guided search, making it suitable for logistics, surveillance, and search-and-rescue tasks where reliable large-scale coordination is essential.

Abstract

Multi-robot path planning is a fundamental yet challenging problem due to its combinatorial complexity and the need to balance global efficiency with fair task allocation among robots. Traditional swarm intelligence methods, although effective on small instances, often converge prematurely and struggle to scale to complex environments. In this work, we present a structure-induced exploration framework that integrates structural priors into the search process of the ant colony optimization (ACO). The approach leverages the spatial distribution of the task to induce a structural prior at initialization, thereby constraining the search space. The pheromone update rule is then designed to emphasize structurally meaningful connections and incorporates a load-aware objective to reconcile the total travel distance with individual robot workload. An explicit overlap suppression strategy further ensures that tasks remain distinct and balanced across the team. The proposed framework was validated on diverse benchmark scenarios covering a wide range of instance sizes and robot team configurations. The results demonstrate consistent improvements in route compactness, stability, and workload distribution compared to representative metaheuristic baselines. Beyond performance gains, the method also provides a scalable and interpretable framework that can be readily applied to logistics, surveillance, and search-and-rescue applications where reliable large-scale coordination is essential.

Figures (6)

• Figure 1: Three stage tour construction on the complete graph of target points: (a) the original spatial distribution of vertices prior to any heuristic; (b) the MST backbone obtained via Kruskal’s algorithm and its Eulerian skeleton produced by a depth first traversal, with provisional edges and nodes highlighted; and (c) the resulting Hamiltonian circuit after ACO refinement, in which redundant backtracks have been pruned to yield a single loop visiting each node exactly once.
• Figure 2: Distributed task allocation for $k$ ant agents: the full set of $n$ target nodes $\{1,2,\dots,n\}$ is partitioned into $k$ disjoint subsets, each of which is assigned to a different ant for parallel path planning.
• Figure 3: Ant transition decision at node $N$: the probability $P_{N\to k}(t)$ of moving to candidate vertex $k\in\{i,j\}$ is computed by normalizing the product of pheromone intensity $\tau_{Nk}(t)$, heuristic desirability $\eta_{Nk}=1/d_{Nk}$, and structural-prior bias $\psi_{Nk}$ over the two options.
• Figure A1: Spatial distribution of robot trajectories for eight different multi-robot routing algorithms on a 200-city, 8-robot instance. From left to right, top to bottom: SINE, ACO, DL-ACO, Genetic Algorithms, AR-ACO, Smooth ACO, IEACO, and Simulated Annealing. SINE demonstrates highly structured and nonoverlapping routes that adhere to the underlying structural backbone. In contrast, other algorithms tend to produce denser, overlapping, or tangled trajectories, which may lead to suboptimal load balancing and increased total travel cost. These differences highlight the effectiveness of structural-prior initialization in producing well organized routes and ensuring clear division of workload among robots.
• Figure A2: Validation on a South Pole coastal sector in 50 nodes. Pink dots denote target sites sampled from the satellite image, while colored polylines represent robot trajectories produced by different algorithms with SINE, ACO, AR-ACO, DL-ACO, IE-ACO. This experiment evaluates transferability under realistic spatial constraints such as irregular terrain boundaries, clustered node distributions, and heterogeneous densities. Compared to other baselines, SINE produces more compact and balanced routes with reduced overlap, demonstrating its robustness in real world geographic scenarios.