Sustainable Task Allocation of Mobile Electric Vehicle Charging Robots via Auction-Driven Deep Reinforcement Learning

Abstract

The rapid proliferation of electric vehicles (EVs) is placing unprecedented pressure on fixed charging infrastructure, particularly in regions with constrained power grids. Mobile charging robots have emerged as a versatile solution, offering scalable deployment and the ability to integrate renewable energy sources dynamically. This paper proposes a novel Auction-based Deep Reinforcement Learning (AD-DRL) framework for the efficient and sustainable coordination of mobile charging robots. The framework synergizes DRL for long term strategic policy optimization with an auction based mechanism for robust, real time task allocation. To address the complexities of robot coordination, we introduce Parameterized Action Q-Learning to manage hybrid action spaces and Reward Neural Networks to balance dynamic multi objective trade offs, supported by a Reverse Validation mechanism to ensure assignment feasibility. Extensive simulations demonstrate that the proposed framework significantly outperforms state-of-the-art baselines, including PPO, DDPG, and metaheuristic approaches. Specifically, our method improves the number of EVs served by 28.5%, increases total energy delivery by 46%, and reduces the average charging cost per EV by 11.1%. These results underscore the potential of the AD-DRL framework as a high performance solution for smart, sustainable, and cost effective EV charging operations in modern urban environments.