Multi-Robot gas-source localization based on reinforcement learning

Multi-robot based gas source localization (GSL) in turbulence dominated airflow environments is addressed. A multi-agent reinforcement learning (RL) algorithm is proposed for training multiple robots to finish the GSL task. To improve searching efficiency, the strategy-sharing based RL algorithm is implemented for the GSL task in three different large-scale advection-diffusion simulated plume environments by using different number of robots. Simulation results show that multiple robots could successfully locate the gas source in turbulence dominated airflow environments with the proposed algorithm; Moreover, the results also demonstrate that the strategy-sharing RL outperforms the RL which does not share strategies.