Connectivity Guaranteed Multi-robot Navigation via Deep Reinforcement Learning

This paper considers the multi-robot navigation problem where the geometric center of a multi-robot team aims to efficiently reach the waypoint without collisions in unknown complex environments while maintaining connectivity during the navigation. A novel Deep Reinforcement Learning (DRL)-based approach is proposed to derive end-to-end policies for the multi-robot navigation problem. In order to guarantee the connectivity during the navigation, a constraint satisfying parametric function (CSPF) is proposed to represent the navigation policy. Virtual policy extended environment (VP2E), an implementation framework of the CSPF is accompanied so as to make CSPF compatible with existing DRL techniques which rely on differentiable parametric functions. Both simulations and real-world experiments of a team of 3 holonomic robots are conducted to verify the effectiveness of the proposed DRL-based navigation method.