Autonomous Agents for Traffic Simulation and Control
Vikram Manikonda, Renato Levy, Goutam Satapathy, David J. Lovell, Peter C. Chang, and Anna Teittinen
Transportation Research Record 1774, pp. 1-10.
ABSTRACT
This paper describes the development of an infrastructure for multiple autonomous agents, with an application to urban traffic signal control. The agent-based infrastructure, named Cybele, allows for distributed computing, inter-agent communication, agent migration, and computational resource allocation. The set of agents that are used to solve the traffic signal control problem are known collectively as DAARTS (Decentralized Adaptive Agents for contRol of Traffic Signals). DAARTS adopts a hierarchical multi agent-based architecture where the lowest level (intersection agents) involves individual intersection-traffic dynamics and phase selection based on ``local'' information, while higher levels take into account the supervisory (network-level) dynamics. The controller design is based on a receding horizon model predictive control approach. Coordination between intersections is achieved in a decentralized manner at the lowest level. The agents are integrated into a simulation testbed with the microsimulator CORSIM, using the DAARTS Simulation Toolkit, or DSTK. This toolkit enables communications between the CORSIM real-time extension, the communications management functions of Cybele, and the traffic control agents. The control process is truly distributed, and each of these components can reside on a different computer. Descriptions of all of the software components are given, and the control algorithm is discussed in detail. Some encouraging results from the simulation of a small network are included. Ongoing and future research activities are discussed.