Informative Trajectory Planning for Air-Ground Cooperative Monitoring of Spatiotemporal Fields

This paper investigates an air-ground cooperative monitoring problem for spatiotemporal fields, such as air pollution, forest fires, oil spills, etc, with unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). To fully exploit complementarities of these heterogeneous vehicles and improve efficiency of the cooperative monitoring, we design a novel cooperation scheme: each UAV is assigned to loiter over and transmit its observations to a pre-allocated UGV, and the UGV provides guidance on informative trajectories for the UAV and aims to reach a target position as fast as possible. Such a scheme brings challenges to informative trajectory planning of the UGVs, lying in the delayed observations from the UAVs and the cumulative information constraint depending on the unknown field. To overcome them, this work proposes a model-free reinforcement learning (RL)-based trajectory planning method to learn continuous policies for the UGVs, where a field estimator is designed for each UGV to recover observability of the field. In addition, we derive model predictive control (MPC)-based trajectory planners for the UAVs with tailored reference positions, where the uncertain tracking errors can be handled by the RL-based method of the UGVs. Thus, a performance coupling problem of the heterogeneous vehicles is tackled. Simulations illustrate the effectiveness of the proposed trajectory planning methods and the efficiency of the air-ground cooperative monitoring scheme.
Note to Practitioners
—This article is motivated by cooperative monitoring tasks with UAVs and UGVs in practical applications, such as environmental monitoring, search and rescue after disasters, etc. Due to the complex dynamics of spatiotemporal fields in these tasks, trajectory planning for the cooperative monitoring system is challenging and requires much computations. To resolve the issues, we propose a novel cooperation scheme in this article, where the large computational capability of the UGVs is utilized to solve a minimum-time trajectory planning problem under a cumulative information constraint, and the UAVs only loiter over and transmit measurements about the field to the UGVs. To achieve this scheme, RL-based and MPC-based trajectory planning methods are proposed for the UGVs and the UAVs, respectively. Simulations have validated the effectiveness of the proposed trajectory planning methods and good performance of the cooperative monitoring system.