AgentSense: LLMs Empower Generalizable and Explainable Web-Based Participatory Urban Sensing

TL;DR

AgentSense tackles the inherent challenges of web-based participatory urban sensing by marrying a classical planner with an LLM-powered multi-agent refinement loop. It introduces a disturbance parser and three coordinated agents (Solver, Eval, Memory) to iteratively adapt task assignments under real-time disturbances while generating natural-language explanations. Across two large mobility datasets and seven disturbance types, AgentSense demonstrates superior adaptivity and interpretability compared to single-agent LLM baselines and traditional planners, with robust performance under budget and feasibility constraints. The framework's zero-shot generalization and transparent reasoning hold significant practical potential for scalable, explainable urban sensing on the web.

Abstract

Web-based participatory urban sensing has emerged as a vital approach for modern urban management by leveraging mobile individuals as distributed sensors. However, existing urban sensing systems struggle with limited generalization across diverse urban scenarios and poor interpretability in decision-making. In this work, we introduce AgentSense, a hybrid, training-free framework that integrates large language models (LLMs) into participatory urban sensing through a multi-agent evolution system. AgentSense initially employs classical planner to generate baseline solutions and then iteratively refines them to adapt sensing task assignments to dynamic urban conditions and heterogeneous worker preferences, while producing natural language explanations that enhance transparency and trust. Extensive experiments across two large-scale mobility datasets and seven types of dynamic disturbances demonstrate that AgentSense offers distinct advantages in adaptivity and explainability over traditional methods. Furthermore, compared to single-agent LLM baselines, our approach outperforms in both performance and robustness, while delivering more reasonable and transparent explanations. These results position AgentSense as a significant advancement towards deploying adaptive and explainable urban sensing systems on the web.

Figures (10)

• Figure 1: (a): Web-based urban sensing under dynamic conditions. (b): AgentSense adaptively refines baseline solutions in response to evolving requirements.
• Figure 2: The AgentSense framework for iterative refinement of urban sensing task solution.
• Figure 3: Eval Agent.
• Figure 4: Memory Agent.
• Figure 5: Iterative improvement process for disturbance "Area blocked of (4, 4)".