Distributed consensus algorithms for merging feature-based maps with limited communication

In this paper we present a solution for merging feature-based maps in a robotic network with limited communication. We consider a team of robots that explore an unknown environment and build local stochastic maps of the explored region. After the exploration has taken place, the robots communicate and build a global map of the environment. This problem has been traditionally addressed using centralized schemes or broadcasting methods. The contribution of this work is the design of a fully distributed approach which is implementable in scenarios with limited communication. Our solution does not rely on a particular communication topology and does not require any central agent, making the system robust to individual failures. Information is exchanged exclusively between neighboring robots in the communication graph. We provide distributed algorithms for solving the three main issues associated to a map merging scenario: establishing a common reference frame, solving the data association, and merging the maps. We also give worst-case performance bounds for computational complexity, memory usage, and communication load. Simulations and real experiments carried out using various vision sensors validate our results.

Research highlights
► Fully distributed algorithms for merging feature-based maps acquired by a robot team.
► We solve the common reference frame, the data association, and the map merging.
► The algorithm is designed for robot networks with limited communication.
► It is robust to changes in the topology and link failures.
► We prove the convergence to the optimal solution.