Communication-efficient planning and mapping for multi-robot exploration in large environments

Abstract

This letter presents a framework for planning and perception for multi-robot exploration in large and unstructured three-dimensional environments. We employ a Gaussian mixture model for global mapping to model complex environment geometries while maintaining a small memory footprint which enables distributed operation with a low volume of communication. We then generate a local occupancy grid for use in planning from the Gaussian mixture model using Monte Carlo ray tracing. Then, a finite-horizon, information-based planner uses this local map and optimizes sequences of observations locally while accounting for the global distribution of information in the robot state space which we model using a library of informative views. Simulation results demonstrate that the proposed system is able to maintain efficiency and completeness in exploration while only requiring a low rate of communication.