Distributed maze exploration using multiple agents and optimal goal assignment

TL;DR

The paper tackles efficient distributed maze exploration under constrained broadcast ranges by introducing CU-LVP, a frontier-exploration method that uses Voronoi partitioning to assign subregions to agents. It blends a wavefront-based goal selection with a cost-utility function $utility(f)=N(u_{mnm}(f))+\lambda \cdot N(u_{jgr}(f))$ and a region-merge mechanism to adapt to communication limits, achieving low computation and communication overhead. Empirical results across 30×30 mazes with varying obstacle densities show CU-LVP often minimizes exploration time and cost while maintaining high map quality, and it attains the highest Composite Index Score among baselines, particularly with more agents in complex mazes. The work demonstrates practical viability for multiagent exploration in real-world settings and points to future enhancements by integrating elements from HEDAC and reinforcement-learning-based approaches for further gains in simpler mazes and different collaboration regimes.

Abstract

Robotic exploration has long captivated researchers aiming to map complex environments efficiently. Techniques such as potential fields and frontier exploration have traditionally been employed in this pursuit, primarily focusing on solitary agents. Recent advancements have shifted towards optimizing exploration efficiency through multiagent systems. However, many existing approaches overlook critical real-world factors, such as broadcast range limitations, communication costs, and coverage overlap. This paper addresses these gaps by proposing a distributed maze exploration strategy (CU-LVP) that assumes constrained broadcast ranges and utilizes Voronoi diagrams for better area partitioning. By adapting traditional multiagent methods to distributed environments with limited broadcast ranges, this study evaluates their performance across diverse maze topologies, demonstrating the efficacy and practical applicability of the proposed method. The code and experimental results supporting this study are available in the following repository: https://github.com/manouslinard/multiagent-exploration/.

Figures (9)

• Figure 1: Hierarchical Structure of the Related Work (Section \ref{['sec:relwork']}). Method in bold (CU-LVP) is our proposed method.
• Figure 2: Wavefront distance calculation for the entire maze. The gray dot indicates the agent's position, whereas each empty cell contains the distances from the agent. The black cells represent obstacles.
• Figure 3: Abstract overview of proposed methodology. Implementation details are provideed in Section \ref{['sec:proposed_method']}.
• Figure 4: Visualization of the initial Voronoi regions for a randomly 30x30 generated maze (top image). Agents within broadcast range (bottom image) have merged their Voronoi regions and exchanged information about their surroundings. Additionally, on the bottom, the red line depicts the wavefront path ($x_1, x_2$) from an agent ($x_0$) to its goal ($f$) within its region. The black cells represent the agents. The colored cells depict the Voronoi regions, with each color representing a different partition, whereas the darker gray cells denote obstacles.
• Figure 5: Copeland Comparison for different $\lambda$ values of CU-LVP across 100 experiments in various maze sizes (namely 15x15, 30x30, and 50x50).