A decentralized intersection management system through collaborative negotiation between smart signals

Abstract

Actuated and pre-timed traffic signal controllers have been beneficial to the improvement of traffic flow in cities and dense urban environments around the world. While these methods have been effective in reducing traffic congestion, recent works have shown that incorporation of reinforcement learning (RL) or other artificial intelligence (AI) based optimization techniques may further improve the performance of traffic signal controllers. This work investigates a novel decentralized traffic signal control structure which encourages cooperative signal behavior via repeated negotiations between neighboring intelligent agents. This method capitalizes on emerging inter-infrastructure communications technologies to exercise ‘system-level’ control over a network of connected signalized intersections. The proposed method was tested in a simplified grid-network of 20 intersections. In this network, static arrivals of $1440\mathrm{ }\text{veh}/\mathrm{l}/\mathrm{h}$ along east-west lanes and $360\mathrm{ }\text{veh}/\mathrm{l}/\mathrm{h}$ along north-south lanes were supplied. In addition, a simulated shift to $720\mathrm{ }\text{veh}/\mathrm{l}/\mathrm{h}$ along east-west lanes and $1080\mathrm{ }\text{veh}/\mathrm{l}/\mathrm{h}$ along north-south lanes was analyzed to provide insights into the presented method’s performance in response to any sudden shifts in traffic patterns. The findings indicated that, when compared to non-negotiating traffic signals, the presented method may improve the service rate of traffic networks under static conditions by $671\mathrm{ }\text{veh}/\mathrm{h}$ on average, reduce emissions by an average of $326\mathrm{ }\text{kg}/\mathrm{h}$ in addition to reducing travel time across a network of intersections. The performance characteristics were captured by a SUMO testbed, and computational efficiency was explored using a suite of simple testbeds developed in MATLAB.

Keywords:

• Decentralization
• distributed control
• intelligent agents
• intelligent control systems
• signalized intersections

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by NSF CIS Award #1538139… and the work is partially supported by the Oak Ridge National Laboratory ASTRO program supported by the US Department of Energy and the SMART initiative of the Vehicle Technologies Office at the US Department of Energy.