Long-Horizon Active SLAM system for multi-agent coordinated exploration

Abstract

Exploring efficiently an unknown environment with several autonomous agents is a challenging task. In this work we propose an multi-agent Active SLAM method that is able to evaluate a long planning horizon of actions and perform exploration while maintaining estimation uncertainties bounded. Candidate actions are generated using a variant of the Rapidly exploring Random Tree approach (RRT*) followed by a joint entropy minimization to select a path. Entropy estimation is performed in two stages, a short horizon evaluation is carried using exhaustive filter updates while entropy in long horizons is approximated considering reductions on predicted loop closures between robot trajectories. We pursue a fully decentralized exploration approach to cope with typical uncertainties in multiagent coordination. We performed simulations for decentralized exploration planning, which is both dynamically adapting to new situations as well as concerning long horizon plans.