Few-Shot Learning for Dynamic Operations of Automated Electric Taxi Fleets under Evolving Charging Infrastructure: A Meta-Deep Reinforcement Learning Approach

TL;DR

This work tackles the challenge of operating Autonomous Electric Taxi fleets under evolving charging infrastructure by introducing GAT-PEARL, a hierarchical meta-reinforcement learning framework that combines graph attention networks for spatial encoding with probabilistic context embeddings for fast, inference-based adaptation. The architecture partitions control into a Central Agent for global coordination and multiple Area Agents for regional, context-aware decision making, all guided by a deterministic heuristic execution layer. Through meta-training across diverse charging layouts and few-shot adaptation to unseen configurations, the approach achieves faster convergence, better generalization, and lower performance variance than baselines in large-scale simulations calibrated with Chengdu data. The results demonstrate the practical value of topology-aware, meta-adaptive fleet control for maintaining service quality and economic efficiency during infrastructure upgrades and disruptions, highlighting implications for planning and operations in evolving urban charging networks.

Abstract

With the rapid expansion of electric vehicles (EVs) and charging infrastructure, the effective management of Autonomous Electric Taxi (AET) fleets faces a critical challenge in environments with dynamic and uncertain charging availability. While most existing research assumes a static charging network, this simplification creates a significant gap between theoretical models and real-world operations. To bridge this gap, we propose GAT-PEARL, a novel meta-reinforcement learning framework that learns an adaptive operational policy. Our approach integrates a graph attention network (GAT) to effectively extract robust spatial representations under infrastructure layouts and model the complex spatiotemporal relationships of the urban environment, and employs probabilistic embeddings for actor-critic reinforcement learning (PEARL) to enable rapid, inference-based adaptation to changes in charging network layouts without retraining. Through extensive simulations on real-world data in Chengdu, China, we demonstrate that GAT-PEARL significantly outperforms conventional reinforcement learning baselines, showing superior generalization to unseen infrastructure layouts and achieving higher overall operational efficiency in dynamic settings.

Figures (22)

• Figure 1: A toy example illustrating the failure of a static policy under evolving infrastructure.
• Figure 2: Hexagonal regions map.
• Figure 3: Temporal hierarchy of strategic periods and operational intervals.
• Figure 4: Categorization of AET movements.
• Figure 5: The hierarchical multi-agent meta-reinforcement learning architecture.