LLM-Assisted Light: Leveraging Large Language Model Capabilities for Human-Mimetic Traffic Signal Control in Complex Urban Environments

TL;DR

This paper tackles urban traffic congestion by integrating Large Language Models into traffic signal control, addressing the adaptability gaps of rule-based and RL-based methods. It introduces LA-Light, a hybrid framework where an LLM selects and uses a suite of perception and decision tools to reason about static and dynamic traffic data, guiding traditional TSC methods and providing transparent justifications. Through SUMO-based simulations on synthetic and real-world networks, LA-Light demonstrates robust performance in normal conditions and rare events, achieving substantial reductions in average travel and waiting times, and improved emergency-vehicle metrics, notably under sensor outages. The work highlights the potential for real-world deployment by offering a standardized interface and explainable decision-making, while outlining future work to reduce latency and incorporate vision-based data for even faster, more reliable control.

Abstract

Traffic congestion in metropolitan areas presents a formidable challenge with far-reaching economic, environmental, and societal ramifications. Therefore, effective congestion management is imperative, with traffic signal control (TSC) systems being pivotal in this endeavor. Conventional TSC systems, designed upon rule-based algorithms or reinforcement learning (RL), frequently exhibit deficiencies in managing the complexities and variabilities of urban traffic flows, constrained by their limited capacity for adaptation to unfamiliar scenarios. In response to these limitations, this work introduces an innovative approach that integrates Large Language Models (LLMs) into TSC, harnessing their advanced reasoning and decision-making faculties. Specifically, a hybrid framework that augments LLMs with a suite of perception and decision-making tools is proposed, facilitating the interrogation of both the static and dynamic traffic information. This design places the LLM at the center of the decision-making process, combining external traffic data with established TSC methods. Moreover, a simulation platform is developed to corroborate the efficacy of the proposed framework. The findings from our simulations attest to the system's adeptness in adjusting to a multiplicity of traffic environments without the need for additional training. Notably, in cases of Sensor Outage (SO), our approach surpasses conventional RL-based systems by reducing the average waiting time by $20.4\%$. This research signifies a notable advance in TSC strategies and paves the way for the integration of LLMs into real-world, dynamic scenarios, highlighting their potential to revolutionize traffic management. The related code is available at https://github.com/Traffic-Alpha/LLM-Assisted-Light.

Figures (11)

• Figure 1: Comparative framework analysis between LA-Light and conventional TSC systems. (a) illustrates a conventional TSC system wherein decisions are made directly by an algorithm that processes environmental inputs. (b) depicts the proposed LA-Light framework, which employs an LLM for the task of traffic signal control. In LA-Light, the LLM begins by selecting the most relevant tools from an enhanced set, including perception tools and decision-making algorithms, to collect and analyze traffic data. It then evaluates the information, adjusting its choice of tools as needed, until a definitive traffic control decision is formulated.
• Figure 2: A standard four-legged road intersection with eight traffic movements and signal phases.
• Figure 3: The LA-Light framework: A schematic representation of the five-step process for integrating LLM in TSC. Step 1 outlines the task planning phase where the LLM defines its role in traffic management. Step 2 involves the selection of appropriate perception and decision-making tools by the LLM. In Step 3, these tools interact with the traffic environment to gather data. Step 4 depicts the analysis of this data by the Decision Unit to inform decision-making. Finally, Step 5 illustrates the implementation of the LLM's decisions and the provision of explanatory feedback for system transparency and validation.
• Figure 4: Examples of tools in the LA-Light framework. Two perception tools, Get_Signal_Phase_Structure and Get_Occupancy, allow users to obtain the traffic signal phase and congestion level of the intersection, respectively. Two decision tools, Get_Auxiliary_Decision provide the decision of the RL-based method, and Justify_Decision_Logic explains the decision according to the current situation of the intersection.
• Figure 5: System Prompt structure within the LA-Light framework. The design incorporates five components: (1) Task Description, detailing the LLM's role in traffic signal management; (2) Tools Synopsis, providing a catalog and description of available traffic control tools; (3) Observations Data, compiling data from tool feedback and chat history of the preceding cycle; (4) Attention Points, emphasizing compliance with traffic regulations and safety guidelines in tool deployment; and (5) Output Format, defining the protocol for the LLM's decision communication to ensure proper tool utilization.