Gaussian Process based Non-myopic Cooperative Exploration of Multiple Robots
Junjie Fu, Guanghui Wen · 2022 IEEE International Conference on Unmanned Systems (ICUS) · 2022
In this paper, we consider the cooperative efficient exploring problem for multi-robot systems in unknown environment. Gaussian process (GP) is employed to build an online environmental model based on the measurements of the robots. Information-theoretic metrics are proposed to facilitate the optimal exploring trajectory planning. A non-myopic model predictive control (MPC) based exploration strategy is designed which considers the effect of multiple future steps. To reduce the computational costs of the centralized MPC optimization problem, sequential greedy strategy is utilized to determine the trajectories of the robots in turn while considering the collision avoidance requirement between the robots. To further reduce the computational cost resulting from the standard GP regression, sparse spectrum Gaussian process (SSGP) is utilized which incurs constant training and prediction costs for the construction of the MPC optimization problem at each time step. Simulation examples are provided which demonstrate the effectiveness of the proposed control strategies.