Research on Informative Path Planning Using Deep Reinforcement learning

Efficient data gathering is important for mapping and surveilling for many applications on the Earth's surface. In large-area monitoring situations, executing a crew of Unmanned Aerial Vehicles (UAVs) provides enhanced spatial coverage and reliability against individual malfunctions. Yet, a major obstacle is developing cooperative path-planning strategies that allow UAVs to accomplish a collaborative mission objective. We present a new Multi-Agent Deep Reinforcement Learning-based (MARL) informative path planning approach for adjustable landscape surveilling using UAV crews. Our approach presents innovative network representations facilitating effective path planning in 3D workspaces. By utilizing a counterfactual baseline, our method effectively resolves the problem of credit assignment to learn cooperative behavior. Our findings demonstrate that enhanced planning performance. Experimental results show that our method outclasses other non-learning-based techniques while generalizing well to various team sizes and communication scenarios.