JADE, TraSMAPI and SUMO: A tool-chain for simulating traffic light control

TL;DR

This paper introduces an open-source tool-chain combining JADE, TraSMAPI, and SUMO to facilitate the development and testing of multi-agent-based traffic light control solutions, demonstrating its practicality through a reinforcement learning example.

Contribution

It presents a novel integrated framework that connects multi-agent systems with traffic simulation tools for research and development in intelligent transportation systems.

Findings

Framework is feasible for experimental traffic control solutions

Reinforcement learning agents successfully integrated within the system

Open-source tool-chain supports diverse agent-based traffic management strategies

Abstract

Increased stress, fuel consumption, air pollution, accidents and delays are some of the consequences of traffic congestion usually incurring in tremendous economic impacts, which society aims to remedy in order to leverage a sustainable development. Recently, unconventional means for modeling and controlling such complex traffic systems relying on multi-agent systems have arisen. This paper contributes to the understanding of such complex and highly dynamic systems by proposing an open-source tool-chain to implement multi-agent-based solutions in traffic and transportation. The proposed approach relies on two very popular tools in both domains, with focus on traffic light control. This tool-chain consists in combining JADE (Java Agent DEvelopment Framework), for the implementation of multi-agent systems, with SUMO (Simulation of Urban MObility), for the microscopic simulation of traffic interactions. TraSMAPI (Traffic Simulation Manager Application Programming Interface) is used to combine JADE and SUMO allowing communication between them. A demonstration of the concept is presented to illustrate the main features of this tool-chain, using Q-Learning as the reinforcement learning method for each traffic light agent in a simulated network. Results demonstrate the feasibility of the proposed framework as a practical means to experiment with different agent-based designs of intelligent transportation solutions.