<p>Multi-agent embodied intelligence has been integrated into warehouse systems for transporting commodities, which can be formulated as the lifelong multi-agent pickup and delivery (LMAPD) problem. However, existing research has not addressed the energy limitations associated with the LMAPD problem, rendering it inapplicable to real-world systems. In this study, we formulate the energy-limited lifelong multi-agent pickup and delivery (EL-MAPD) problem, in which each agent consumes or recharges a specific amount of energy for its actions. Furthermore, we theoretically define a realistic subclass of EL-MAPD instances known as well-formed EL-MAPD instances. This subclass ensures the existence of feasible solutions, preventing scenarios in which agents run out of energy or collide with each other. To obtain a high-quality feasible solution under the framework of concurrent planning and execution, we propose the fallback priority planning (FPP) algorithm. The FPP algorithm employs a fallback mechanism to ensure the acquisition of a feasible action in each planning episode. Experimental results demonstrate that the FPP algorithm effectively balances throughput and energy consumption. Furthermore, to enhance the warehouse system’s throughput in real-world scenarios, our findings suggest decreasing the planning time limit for each planning episode, increasing the number of agents, and implementing a recharge strategy wherein agents complete the recharging process only when their energy levels are full.</p>