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

2026-05-31

AI Summary

Existing approaches struggle to simultaneously achieve interpretability, behavioral plausibility, population-level distribution alignment, and computational efficiency in human mobility trajectory generation. This work proposes the first agent-based, self-evolving heuristic system, wherein large language model (LLM) agents perform bias diagnosis and failure analysis on a validation set to iteratively refine initial heuristic rules. An evolutionary memory mechanism enables continuous improvement through accumulated experience. By integrating behavioral heuristic modeling, agent-driven self-diagnosis, rule evolution, and memory retention, the method substantially outperforms current state-of-the-art deep generative and LLM-based approaches on benchmarks from Singapore and Montreal. It excels in individual trajectory fidelity, population distribution alignment, and behavioral realism, while maintaining high inference efficiency and strong interpretability.

Abstract

Human mobility generation aims to synthesize realistic trip chains for target populations based on individual features. Existing paradigms, including deep generative models, LLM-based methods, and traditional heuristics, struggle to satisfy the complex demands of this task while simultaneously maintaining interpretability, behavioral plausibility, population-level distributional alignment, and inference efficiency. To bridge this gap, we introduce MobEvolve, the first agentic self-evolving heuristic framework for human mobility generation. MobEvolve initializes a behavior-inspired heuristic system and employs an LLM agent to iteratively evolve its internal logic. By diagnosing empirical misalignments and failure cases on a validation set, the agent proposes targeted updates and accumulates evolution memory for cumulative self-improvement. Extensive evaluations on the Singapore and Montreal benchmarks demonstrate that MobEvolve significantly outperforms state-of-the-art deep generative and LLM-based methods in individual trajectory fidelity, population-level distribution alignment, and behavioral plausibility, while preserving interpretability and high inference efficiency.

Problem

• human mobility generation
• interpretability
• behavioral plausibility
• distributional alignment
• inference efficiency

Innovation

• agentic self-evolution
• interpretable mobility generation
• heuristic system
• LLM agent
• behavioral plausibility