Defesa – Dissertação de Madson Rodrigues Lemos

Aluno(a): Madson Rodrigues Lemos
Orientador(a): Mariana Luderitz Kolberg Fernandes

Título: Loop-Aware Graph-Based Coordination for Multi-Robot Exploration
Linha de Pesquisa: Planejamento, Sistemas Multiagentes e Robótica

Data: 23/06/2026
Hora: 10:00
Local: Esta banca ocorrerá de forma remota. Acesso público disponibilizado pelo link https://mconf.ufrgs.br/webconf/00132229.

Banca Examinadora:
-Edison Pignaton de Freitas (UFRGS)
-Dante Augusto Couto Barone (UFRGS)
-Mathias Fassini Mantelli (Sereact – Stuttgart)

Presidente da Banca: Mariana Luderitz Kolberg Fernandes

Resumo: Autonomous exploration of unknown environments by teams of mobile robots is a central capability for applications in which human presence is dangerous, costly, or impractical, including disaster response, environmental monitoring, critical-infrastructure inspection, and remote industrial inspection. Multi-robot teams can reduce mission time and improve robustness, but these advantages depend on coordination mechanisms that distribute robots efficiently while preserving the quality of the map produced by the underlying SLAM system. Most coordinated exploration methods emphasize geometric coverage, information gain, or task-allocation efficiency. These objectives are important, but they often treat the map as a passive product of exploration and do not explicitly account for the way frontier choices influence odometric drift, loop-closure opportunities, and map consistency. This dissertation presents MLAEG, a Multi-robot Loop-Aware Exploration Graph framework for coordinated exploration. MLAEG extends the single-robot Loop-Aware Exploration Graph (LAEG) to multi-robot teams by combining local loop-aware topological reasoning with a lightweight centralized coordination layer. Each robot builds a local graph from its occupancy grid, classifies graph nodes as outer frontiers, inner frontiers, connectors, and robot nodes, and scores candidate frontiers according to loop-closure potential. A centralized coordinator receives compact information from these LAEG graphs, including frontier nodes, local scores, frontier types, and robot pose, and assigns goals using a decision rule that combines local attractiveness with global penalties for robot proximity, assignment conflict, short-term revisitation, and recent target oscillation. Long-term visitation remains encoded in each robot’s local LAEG as visited nodes and edges, while the coordinator maintains a short-term memory of recently visited or recently assigned positions. This dual revisit-control mechanism suppresses immediate redundant outer-frontier assignments while preserving inner-frontier candidates that may enable cross-robot loop closure. The method was implemented in ROS1 Noetic and evaluated with four simulated Pioneer 3-DX robots in five environments: Corridors, Shop, Office, Sparse, and Castle. MLAEG was compared with four coordinated multi-robot exploration techniques: Utility-based Frontier Coordination (UFC), Hungarian Frontier Allocation (HFA), V-R-MATD3, and Market-based Frontier-RRT (MB-FRRT). For each method-environment pair, five independent trials were executed under the same experimental protocol. Rather than proposing a new multi-robot SLAM back end, this dissertation evaluates how loop-aware exploration coordination influences map consistency and exploration efficiency under a shared local mapping and occupancy-grid merging pipeline. The evaluation considers exploration time, known-map ratio, trajectory redundancy, path length, obstacle matching ratio (OMR), map matching ratio (MMR), computational cost, and trial-to-trial variability. Using medians as the main statistic, MLAEG obtains the best macro-average exploration time, known-map ratio, path length, revisit ratio, OMR, and MMR. The strongest effect is the reduction of measured cell-level trajectory revisiting while maintaining compact team trajectories and high map agreement. The results support the thesis that loop-aware frontier assignment improves coordinated exploration by balancing coverage, spatial dispersion, and structural map consistency.

Palavras-Chave: Multi-Robot Exploration. Autonomous Exploration. Loop Closure. Topological Graphs. Frontier Assignment. Mobile Robotics.