A deep reinforcement learning approach to preserve connectivity for multi-robot systems

Abstract

Multi-robot systems have been extensively studied for potential values in various areas and the connectivity within mobile robots plays an important role in many coordination and cooperative applications. In this paper, we adopt a DDPG-based learning framework to address the connectivity preservation problem for multi-robot system. In the implemented framework, a multi-robot simulation environment is developed to provide states and reward feedbacks to the learning agent based on a actor-critic architecture. The DDPG-based agent applies fully connected neural networks to parameterize a Q-function and a policy function. The adopted learning framework and implemented components (the networks and the simulator) can successfully solve the connectivity preservation problem and they are validated by a set of simulation experiments.