Title: MobEvolve: An Agentic Self-Evolving Heuristic System for Interpretable Human Mobility Generation

Abstract: The objective of human mobility generation is to create realistic trip sequences for specific demographic groups using individual characteristics. Current approaches, such as deep generative models, methods based on Large Language Models (LLMs), and conventional heuristics, often fail to meet the intricate requirements of this domain. Specifically, they struggle to balance interpretability, behavioral realism, alignment with population-level distributions, and computational efficiency. To address these limitations, we present MobEvolve, a novel agentic self-evolving heuristic framework designed for human mobility synthesis. This system begins with a heuristic model inspired by human behavior, which is then iteratively refined by an LLM agent. The agent identifies discrepancies and failure instances within a validation set, using these insights to suggest precise adjustments and build an evolving memory bank that facilitates continuous improvement. Comprehensive testing on the Singapore and Montreal datasets reveals that MobEvolve surpasses existing state-of-the-art deep generative and LLM-based techniques. It achieves superior results in maintaining individual trajectory accuracy, aligning with population distributions, and ensuring behavioral plausibility, all while retaining full interpretability and rapid inference speeds.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC