CLMPO-EC: A Lightweight Multi-UAV Multiarea Coverage Path Planning Method Using Deep Reinforcement Learning

doi:10.1109/JIOT.2026.3659864

CLMPO-EC: A Lightweight Multi-UAV Multiarea Coverage Path Planning Method Using Deep Reinforcement Learning

2026-02-03

0

PRE

AI

DOI：10.1109/JIOT.2026.3659864

Original

Original request for help

Save

Abstract

En 中文

Unmanned aerial vehicles (UAVs), as an aerial extension of the Internet of Things (IoT) sensing layer, have played an increasingly important role in applications such as environmental monitoring, disaster assessment, and precision agriculture. These tasks can be uniformly abstracted as coverage path planning (CPP), which aims to achieve efficient scanning and surveying while ensuring complete coverage of single or multiple disconnected regions. While single-region CPP has been extensively studied, in multiregion settings existing methods often rely on predefined coverage patterns to guarantee completeness, which to some extent limits their flexibility. Meanwhile, constraints on UAV energy and onboard computation impose higher performance requirements on planning methods. To address these challenges, this article targets an energy-constrained multi-UAV cooperative scenario and proposes a cross-layer, energy-constrained path optimization framework based on multiagent reinforcement learning (CLMPO-EC). Specifically, the framework organizes the overall task into two layers—CPP and multiagent path planning (MAPP)—and, on this basis, integrates back-and-forth path planning (BFP) with multiagent reinforcement learning (MARL) under a centralized training and distributed execution paradigm to construct a unified, interactive, and structured environmental model. CLMPO-EC further introduces a lightweight cross-layer connection network that propagates raw state information to higher layers to enhance learning efficiency. In addition, building on BFP, an entrance—exit exploration factor is proposed to dynamically adjust the exploration probability of regional entrances and exits in CPP according to the training phase and batch, thereby improving the efficiency of searching for optimal solutions. Theoretical analysis and experimental results demonstrate that the proposed method achieves superior performance in terms of optimality and efficiency.

Keywords:

Coverage path planning (CPP)

deep reinforcement learning (DRL)

Internet of Things (IoT)

multiple regions

multiple unmanned aerial vehicles (UAVs)

CLMPO-EC: A Lightweight Multi-UAV Multiarea Coverage Path Planning Method Using Deep Reinforcement Learning

Abstract

Journal

Organization