TrafficClaw: Generalizable Urban Traffic Control via Unified Physical Environment Modeling

2026-04-19

AI Summary

This work addresses the fragmentation of existing urban traffic control systems—such as traffic signals, highway networks, and public and taxi services—which operate in isolation and lack a unified modeling framework, thereby hindering the capture of coupled dynamics and cross-task generalization. To overcome this limitation, the authors propose the first unified physical environment framework for urban traffic control, integrating heterogeneous subsystems into a shared dynamical system that enables closed-loop interaction and system-level coordination. Built upon this foundation, they deploy a large language model agent endowed with spatiotemporal reasoning and reusable procedural memory, trained via a multi-stage strategy combining supervised pretraining and system-level reinforcement learning. Experiments demonstrate that the approach exhibits strong robustness, transferability, and system-awareness across unseen scenarios, dynamic conditions, and task configurations, significantly outperforming existing specialized methods.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Multiagent Systems: Multiagent Learning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Urban traffic control is a system-level coordination problem spanning heterogeneous subsystems, including traffic signals, freeways, public transit, and taxi services. Existing optimization-based, reinforcement learning (RL), and emerging LLM-based approaches are largely designed for isolated tasks, limiting both cross-task generalization and the ability to capture coupled physical dynamics across subsystems. We argue that effective system-level control requires a unified physical environment in which subsystems share infrastructure, mobility demand, and spatiotemporal constraints, allowing local interventions to propagate through the network. To this end, we propose TrafficClaw, a framework for general urban traffic control built upon a unified runtime environment. TrafficClaw integrates heterogeneous subsystems into a shared dynamical system, enabling explicit modeling of cross-subsystem interactions and closed-loop agent-environment feedback. Within this environment, we develop an LLM agent with executable spatiotemporal reasoning and reusable procedural memory, supporting unified diagnostics across subsystems and continual strategy refinement. Furthermore, we introduce a multi-stage training pipeline with supervised initialization and agentic RL with system-level optimization, further enabling coordinated and system-aware performance. Experiments demonstrate that TrafficClaw achieves robust, transferable, and system-aware performance across unseen traffic scenarios, dynamics, and task configurations. Our project is available at https://github.com/usail-hkust/TrafficClaw.

Problem

• urban traffic control
• heterogeneous subsystems
• cross-task generalization
• coupled physical dynamics
• system-level coordination

Innovation

• unified physical environment
• cross-subsystem interaction
• LLM agent with spatiotemporal reasoning
• system-level traffic control
• multi-stage agentic training