Bridging Policy and Real-World Dynamics: LLM-Augmented Rebalancing for Shared Micromobility Systems
Heng Tan, Hua Yan, Yu Yang
AI summary
Problem
Existing rebalancing methods either ignore real-world emergent events or sacrifice normal operational performance for robustness, leaving operators without a scalable way to adapt to sudden demand surges, vehicle outages, or regulatory changes.
Approach
The AMPLIFY framework decouples a baseline rebalancing policy from an LLM-based adaptation agent that ingests system context and emergent conditions to iteratively refine strategies via self-reflection, enabling real-time, rule-free adjustments.
Key results
- Real-time adaptation to emergent scenarios without policy retraining
- Improved demand satisfaction and system revenue over baselines on Chicago e-scooter data
- Grounding the LLM in baseline strategies enhances adaptation stability and feasibility
- Self-reflection loop effectively corrects infeasible transfers and multi-goal misalignments
Why it matters
Enables urban mobility operators and city regulators to dynamically optimize shared micromobility fleets under uncertainty with minimal computational overhead and no scenario-specific retraining.
Abstract
Shared micromobility services such as e-scooters and bikes have become an integral part of urban trans- portation, yet their efficiency critically depends on effective vehicle rebalancing. Existing methods either optimize for av- erage demand patterns or employ robust optimization and reinforcement learning to handle predefined uncertainties. How- ever, these approaches overlook emergent events (e.g., demand surges, vehicle outages, regulatory interventions) or sacrifice performance in normal conditions. We introduce AMPLIFY, an LLM-augmented policy adaptation framework for shared micromobility rebalancing. The framework combines a baseline rebalancing module with an LLM-based adaptation module that adjusts strategies in real time under emergent scenarios. The adaptation module ingests system context, demand predic- tions, and baseline strategies, and refines adjustments through self-reflection. Evaluations on real-world e-scooter data from Chicago show that our approach improves demand satisfaction and system revenue compared to baseline policies, highlighting the potential of LLM-driven adaptation as a flexible solution for managing uncertainty in micromobility systems.