Schedule-Driven Coordination for Real-Time Traffic Network Control

Abstract

Real-time optimization of the dynamic flow of vehicle traffic through a network of signalized intersections is an important practical problem. In this paper, we take a decentralized, schedule-driven coordination approach to address the challenge of achieving scalable network-wide optimization. To be locally effective, each intersection is controlled independently by an on-line scheduling agent. At each decision point, an agent constructs a schedule that optimizes movement of the observable traffic through the intersection, and uses this schedule to determine the best control action to take over the current look-ahead horizon. Decentralized coordination mechanisms, limited to interaction among direct neighbors to ensure scalability, are then layered on top of these asynchronously operating scheduling agents to promote overall performance. As a basic protocol, each agent queries for newly planned output flows from its upstream neighbors to obtain an optimistic projection of future demand. This projection may incorporate non-local influence from indirect neighbors depending on horizon length. Two additional mechanisms are then introduced to dampen ``nervousness'' and dynamic instability in the network, by adjusting locally determined schedules to better align with those of neighbors. We present simulation results on two traffic networks of tightly-coupled intersections that demonstrate the ability of our approach to establish traffic flows with lower average vehicle wait times than both a simple isolated control strategy and other contemporary coordinated control strategies that use moving average forecast or traditional offset calculation.

Learning Model Parameters for Decentralized Schedule-Driven Traffic Control

“…Although several works in RL has already made progress in traffic signal control problems (Wiering 2000; Kuyer et al 2008;Bazzan and Klügl 2014), they are often slow to converge and difficult to apply under the real-time setting if traffic flows vary frequently. Moreover, if we consider a more realistic setting in which external information is provided solely through local sensors (the most common being physically nearby inroadway induction loop detectors or cameras) rather than relying on global state or vehicle-based representation, the online planning approach (Xie, Smith, and Barlow 2012) is viewed as a recent state of the art (Covell, Baluja, and Sukthankar 2015). In this sense, learning model parameters for planning is a more reasonable solution for realistic traffic signal control systems.…”

“…However, this leads to an increase in complexity as the graph becomes larger. In the field of planning, exchanging information to extend the horizon is considered in (Sen and Head 1997;Gartner, Pooran, and Andrews 2002;Xie, Smith, and Barlow 2012) as a way to accommodate non-local information.…”

“…Recent development of a decentralized online planning approach to the traffic signal control problem has achieved significant traffic flow efficiency improvements through real-time, distributed generation of longhorizon, signal timing plans. (Xie, Smith, and Barlow 2012;Smith et al 2013) The key idea behind this schedule-driven approach is to formulate the intersection scheduling problem as a single machine scheduling problem, where input jobs are represented as sequences of spatially proximate vehicle clusters (approaching platoons, queues).This aggregate representation enables efficient generation of plans with longer horizons that incorporate multi-hop traffic flow information, Copyright c 2020, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.…”

Embedding Automated Planning within Urban Traffic Management Operations

“…While there are several examples of the application of general AI techniques to road traffic monitoring and management (Various 2007;Miles and Walker 2006), the trial of UTM systems embodying an AI planning engine within a real urban traffic management centre, with an evaluation performed by transport operators and technology developers, is novel. On the scheduling side, however, the SURTRAC project utilises a distributed scheduling system which controls traffic signals in urban areas (Xie, Smith, and Barlow 2012). In SURTRAC, each intersection is controlled by a scheduling agent that communicates with connected neighbours to predict future traffic demand, and to minimise predicted vehicles waiting time at the traffic signal.…”

Approximate Gradient Descent Convergence Dynamics for Adaptive Control on Heterogeneous Networks

“…The computational complexity of the scheduling problem on large-scale road networks has motivated the search for efficient decentralized algorithms only requiring local knowledge of network properties and efficient in the absence of coordination. Such decentralized approaches have proven quite efficient in practice (Smith et al 2013;Xie, Smith, and Barlow 2012), have connections with fluid dynamic models (Brett et al 2016), and are amenable to agent-based learning methods such as reinforcement learning (Richter, Aberdeen, and Yu 2007).…”