Proactive Charging Strategy-Based Efficiency Optimization for Multi-UAV Collaborative Planning in Large-Scale Open Environments

Abstract

The limited battery capacity of UAVs poses a fundamental constraint on the sustainability and scalability of multi-UAV cooperation in large-scale open environments. A promising solution is to incorporate battery charging actions into the mission planning phase. However, many charging-aware planners still assume full recharging or insert charging after routing, which may miss routing–charging trade-offs under nonlinear battery charging dynamics. This paper presents a collaborative route planning framework for autonomous UAV swarms that integrates a proactive charging strategy. Specifically, physical charging stations are expanded into a set of extended charging points, each associated with a predefined target state of charge (SoC), and charging decisions are embedded into the planning process so that target-SoC selection is handled as a compact discrete routing decision, formulating the Proactive Charging Strategy based Vehicle Routing Problem (PCS-VRP), which captures the intrinsic coupling between task execution and energy recharging. To efficiently obtain high-quality solutions to the PCS-VRP, a Deep Reinforcement Learning based Route Planning (DRL-RP) algorithm is developed, employing a Transformer architecture with a heterogeneous attention mechanism to model interactions among UAV states, task waypoints, and extended charging points during sequential route construction. Simulation and field experiments show that DRL-RP achieves the shortest total mission time among all compared methods, reducing total mission time by over 10% in large-scale simulations and by 11.39%, 27.19%, and 40.52% in three real-world visual coverage trials relative to the Gurobi baseline. Note to Practitioners—This paper addresses the practical challenge of sustaining multi-UAV operations in large, open environments under endurance constraints. Conventional planning typically optimizes flight routes first and then tunes charging strategies, which overlooks the tight coupling between routing and charging decisions. In contrast, charging decisions are integrated directly into collaborative planning by formulating the PCS-VRP and solving it with DRL-RP, built on a Transformer architecture with heterogeneous attention. Real-world visual coverage experiments demonstrate that collaborative planning with the proactive charging strategy significantly improves mission efficiency. The method can extended to energy-constrained multi-robot systems across diverse scenarios, including infrastructure inspection, emergency response, and smart logistics.