Autonomous Multi-robot Exploration with Ground Vehicles in DARPA Subterranean Challenge Finals

Autonomous navigation and multi-robot exploration framework for ground robots are key parts of the robotic system deployed by the team CTU-CRAS-NORLAB in the final event of the Subterranean (SubT) Challenge organized by the Defense Advanced Research Projects Agency (DARPA) in 2021. The SubT Challenge aimed to advance technologies related to search-and-rescue missions with multi-robot systems in underground environments where communication is unavailable and global navigation satellite systems are denied. This field report describes the developed multi- robot exploration framework focusing on planning, exploration, and traversability estimation for a heterogeneous team of ground vehicles in large-scale rough terrains and multi-robot coordination with limited communication. The developed method employs a dense local mapping for precise traversability estimation combined with a sparse topometrical map shareable between multiple robots and is thus used in the decision-making of the exploration strategy. The topometrical map is designed to support the decentralized coordination of heterogeneous teams of robots, which is demonstrated by deploying the developed framework in the SubT competitions. The framework has been employed in the Virtual track for the full autonomous control of the ground robots, where our team scored second. Besides, in the Systems track, a human supervisor exploited autonomous behaviors provided by the proposed framework to control a heterogeneous team of six ground robots. We report on real-world experimental results from the deployment of the SubT Challenge. Furthermore, we present results from the post-event testing of the SubT Challenge, where three quadruped robots controlled by the framework explored over five hundred meters fully autonomously.