ORBITER Improves Last-Mile Delivery with Conflict-Aware Agentic Decisions

Last-mile delivery faces challenges in dynamically assigning orders to couriers while considering complex spatial and temporal factors. While recent learning-based approaches predict courier sequences, they often lack transparency in their decision-making process. This new research introduces ORBITER, an agentic system designed to enhance next-order decision-making in last-mile logistics.ORBITER structures courier service around "decision points," each capturing the courier's current state and available orders, highlighting local trade-offs. It uses a multi-component system: fixed proposers suggest candidates, an LLM gathers evidence on leading alternatives using task-specific tools, and an independent critic verifies the final decision against the gathered evidence. This conflict-aware approach aims for more reliable and explainable decisions.Evaluations across four cities demonstrate ORBITER's effectiveness, showing an average performance improvement of up to 9.2% over current state-of-the-art baselines. This suggests a significant step forward in optimizing complex last-mile operations through advanced AI agents.