The paper “Design for One, Deploy for Many: Navigating Tree Mazes with Multiple Agents” was selected as a Best Paper Award Finalist at the 5th IEEE International Symposium on Multi-Robot & Multi-Agent Systems (MRS 2025), held December 4–5, 2025 at the Singapore University of Technology and Design (SUTD).

Resulting from a joint effort of OpenSwarm partner Roderich Groß and researchers Jahir Argote-Gerald, Genki Miyauchi, Julian Rau, Paul Trodden, the work was presented on-site by Julian Rau, who received the finalist certificate on behalf of all the authors.

The paper addresses how groups of robots can coordinate effectively in unknown maze-like, confined environments, such as cave and pipe networks, where communication is limited and congestion can quickly become a bottleneck.

 

Research Highlights

The authors propose a distributed multi-agent maze traversal approach for environments that can be represented as tree graphs (acyclic graphs). The method uses a leader-switching mechanism: one robot (“the head”) temporarily assumes the head role and runs a standard single-agent maze solver, while the remaining robots form a follower structure. When the head’s intended next move is blocked by another robot, the head role transfers locally, ensuring the group progresses without collisions while preserving the path that a single robot running the single-agent maze solver would have taken.

The approach was extensively validated, including simulations with up to 300 agents and real-robot experiments with up to 20 Pi-puck robots, demonstrating feasibility and strong performance compared to baseline strategies.