Multi-Agent Pathfinding with TurtleBot 4

Contributors

Overview

Multi-Agent Path Finding (MAPF) focuses on computing collision-free paths for a group of agents. Traditionally, MAPF research has been largely theoretical, often relying on unrealistic assumptions—such as agents operating in a simplified grid world with perfect sensing, communication, and execution. Benchmark datasets and performance evaluations have almost exclusively been conducted under these idealised conditions, overlooking the many complexities of real robot operation, including localisation errors, actuation delays, communication latency, and imperfect motion control. As a result, there is often a significant gap between the throughput reported by MAPF solvers in simulation and the actual throughput achievable with physical robots.

This project aims to bridge that gap by developing a framework and interface that can translate the discrete actions produced by MAPF planners into executable commands for real robots. Using a fleet of TurtleBots, we conduct experiments to validate MAPF solutions in realistic settings. This not only provides a practical testing environment for MAPF planners but also offers valuable insights into the discrepancies between theoretical performance and real-world execution, helping inform the design of more robust and deployment-ready MAPF algorithms.

Summer research 2025

We’re looking for students to take part in Monash summer vacation research scholarships 2025. You’ll spend twelve weeks working with the team to develop systems for real-world multi-agent robotics.

• Applications open: Monday 5 August 2025 (12AM AEST)
• Applications close: Friday 29 August 2025 (5PM AEST)
• Notice of outcome: Mid to late October

If you’re interested, apply here.