The goal of this project was to develop an algorithm to coordinate a team of multiple robots to explore unknown subterranean environments. The objective is to reach complete coverage of an area of interest, while minimizing the total time required to complete the mission. Working inside a simulated environment, the robots are tasked to collaboratively construct a global occupancy map by means of LiDAR sensors.

To achieve this goal, I implemented a modular simulation environment in C++ within the Robot Operating System (ROS) framework, where multiple planning solutions could be tested and benchmarked. I then implemented a planner employing a decentralized coordination algorithm based on Monte Carlo Tree Search (MCTS). The resulting pipeline performs better than other planning solutions in real-world maps from the DARPA SubT challenge, although a more thorough quantitative evaluation is still needed.

The project received a very good evaluation by the supervisors, and is being prepared for peer-reviewed publication. The thesis report and source code will be made publicly available.