CoMAL: Collaborative Multi-Agent Large Language Models for Mixed-Autonomy Traffic (2410.14368v2)

Abstract: The integration of autonomous vehicles into urban traffic has great potential to improve efficiency by reducing congestion and optimizing traffic flow systematically. In this paper, we introduce CoMAL (Collaborative Multi-Agent LLMs), a framework designed to address the mixed-autonomy traffic problem by collaboration among autonomous vehicles to optimize traffic flow. CoMAL is built upon LLMs, operating in an interactive traffic simulation environment. It utilizes a Perception Module to observe surrounding agents and a Memory Module to store strategies for each agent. The overall workflow includes a Collaboration Module that encourages autonomous vehicles to discuss the effective strategy and allocate roles, a reasoning engine to determine optimal behaviors based on assigned roles, and an Execution Module that controls vehicle actions using a hybrid approach combining rule-based models. Experimental results demonstrate that CoMAL achieves superior performance on the Flow benchmark. Additionally, we evaluate the impact of different LLMs and compare our framework with reinforcement learning approaches. It highlights the strong cooperative capability of LLM agents and presents a promising solution to the mixed-autonomy traffic challenge. The code is available at https://github.com/Hyan-Yao/CoMAL.

• The paper introduces CoMAL, a framework that leverages large language models to optimize traffic flow through collaborative multi-agent decision-making.
• It details three core modules—perception, memory, and collaboration—that convert environmental data into actionable strategies for autonomous vehicles.
• Experimental evaluations on the Flow benchmark show increased vehicle speeds and reduced speed variance, highlighting CoMAL’s competitive edge over traditional methods.

Collaborative Multi-Agent LLMs for Mixed-Autonomy Traffic

The paper "CoMAL: Collaborative Multi-Agent LLMs for Mixed-Autonomy Traffic" presents a framework designed to enhance traffic efficiency through the integration of autonomous vehicles using LLMs. The primary goal of CoMAL is to optimize mixed-autonomy traffic systems by facilitating collaboration among autonomous vehicles, thereby improving traffic flow and reducing congestion.

Technical Overview

CoMAL operates within an interactive traffic simulation environment, utilizing several modules to enhance decision-making and cooperation among autonomous vehicles. The framework consists of three primary components:

1. Perception Module: Captures and encodes environmental data into textual descriptions, enabling LLMs to comprehend traffic scenarios effectively.
2. Memory Module: Stores historical driving experiences and instructions, aiding agents in leveraging past knowledge to enhance decision-making processes.
3. Collaboration Module: Facilitates the exchange of information and strategies among multiple agents through a shared message pool, guiding the allocation of tasks and role designation.

Using these modules, CoMAL engages LLMs to formulate strategies and plans in response to dynamic traffic conditions. Each autonomous vehicle, equipped with LLM capabilities, interacts with others to create a collaborative approach aimed at mitigating traffic disturbances like stop-and-go waves.

Experimental Evaluation

The evaluation of CoMAL was conducted using the Flow benchmark, which simulates various traffic scenarios such as Ring, Figure Eight, and Merge. In these setups, the performance of CoMAL was assessed in terms of average velocity and driving smoothness, presenting significant improvements over human-driven setups.

• Average Vehicle Speed: CoMAL exhibited superior traffic flow performance, increasing average vehicle speed across various scenarios.
• Traffic Stability: The standard deviation of vehicle speed under CoMAL decreased, indicating smoother and more stable traffic flow.

The paper also compared CoMAL's performance with reinforcement learning (RL) approaches, revealing competitive results in mixed-autonomy traffic management, particularly in scenarios requiring global collaboration among multiple vehicles.

Implications and Future Directions

The integration of LLMs in the CoMAL framework underscores the potential of knowledge-driven approaches over traditional data-driven methods, particularly in handling complex and unpredictable traffic dynamics. The paper highlights several promising areas for future research:

• Extended Multi-Agent Collaboration: Further exploration of extensive agent collaboration may yield emergent cooperative behaviors similar to those found in RL methods.
• Hybrid Approaches: The combination of RL and LLMs may offer enhanced performance, leveraging the strengths of both methodologies in diverse traffic scenarios.
• Real-World Applications: Transitioning from simulation to real-world traffic systems will be crucial in demonstrating the practical benefits of such frameworks.

In conclusion, CoMAL represents a significant step towards efficient mixed-autonomy traffic systems by exploiting the cooperative capabilities of LLMs. While the framework shows potential in improving traffic dynamics, ongoing research and development will be critical in overcoming current limitations and achieving full-scale implementation in urban traffic environments.