Multi-Agent Robust and Scalable Cooperative Indoor Mapping (MARSCIM)

The problem of sending a team of robots into an unknown environment to create a map is one of the most studied problems in robotics, and is referred to in the literature as Simultaneous Localization and Mapping (SLAM) or Distributed SLAM (DSLAM) when there are multiple robots. Here we propose to extend existing DSLAM capabilities, with robust and scalable algorithms to support increased cooperation and collaboration between teams of unmanned robotic aerial scouts engaged in tunnel mapping operations. Central to our approach is a set of message passing algorithms for information sharing, frontier expansion, and task allocation that are designed explicitly to address the challenges associated with real world robotics. Specifically, our algorithms are designed to operate under uncertainty and to be highly robust to ongoing communication, robotic, and sensor failures. During mission operations a parameter tuning mechanism, which was trained prior to mission execution using machine learning methods, is used to analyze observed information about the environment, and dynamically adjust algorithm parameter settings in order to maximize performance. In this way, the system continually adapts to match the characteristics of the environment in which it operates. Lastly, we propose a rigorous and thorough simulation-based evaluation of the developed algorithms, in order to prove their efficiency and robustness, and to prepare for transition to real-world robotics.