Source-linked AI summary
Distributed Traffic Signal Control for Maximum Network Throughput
Tichakorn Wongpiromsarn, Tawit Uthaicharoenpong, Yu Wang, Emilio Frazzoli, Danwei Wang
TL;DR
Traffic signal control needs strategies that scale beyond small networks while maintaining satisfactory performance. This paper adapts backpressure routing into independently operating local controllers, formally proving maximum network throughput and reporting significantly better performance than SCATS.
Problem
Existing optimization-based traffic signal control approaches face scalability issues, motivating strategies that scale with network size and require minimal tuning.
Method
The paper adapts backpressure routing to construct independently implemented local controllers that use junction-local information without traffic arrival-rate knowledge.
Results
The algorithm is formally proved to achieve maximum network throughput, and simulations show significantly better performance than SCATS.
Takeaways & Limitations
Distributed traffic signal control can achieve maximum network throughput without centralized construction or a global road-network view.
Takeaways & Limitations
Future work identifies fairness constraints, driver-response delays from nonperiodic phase switching, and coordination such as green waves as unresolved issues.
Abstract
from arXiv · showhide
We propose a distributed algorithm for controlling traffic signals. Our algorithm is adapted from backpressure routing, which has been mainly applied to communication and power networks. We formally prove that our algorithm ensures global optimality as it leads to maximum network throughput even though the controller is constructed and implemented in a completely distributed manner. Simulation results show that our algorithm significantly outperforms SCATS, an adaptive traffic signal control system that is being used in many cities.
I. INTRODUCTION
The paper targets scalable traffic signal control by adapting backpressure routing into a distributed strategy that seeks maximum throughput without arrival-rate information. It contrasts this goal with existing approaches’ scalability and tuning challenges.
- SCATS uses calibrated parameters, critical-junction computations, prespecified split plans, and voting to control signal timing.
- Existing optimization and control-theoretic traffic-signal approaches face scalability issues that limit their application to relatively small networks.
- The proposed strategy aims to require minimal tuning, scale with road-network size, and maintain satisfactory performance.
- Backpressure routing motivates the approach because it achieves maximum network throughput without requiring traffic arrival-rate knowledge.
- The paper adapts backpressure routing to traffic signals, formally proves maximum throughput under independent junction control, and reports significant improvement over SCATS.
II. PRELIMINARIES
The preliminaries define network stability, admissible arrival rates, and the capacity region used to characterize maximum throughput. They also introduce Lyapunov drift as a sufficient tool for proving strong stability.
- The network model uses directed links, node queues, slotted time, exogenous arrivals, and departures only when objects reach their destinations.
- Strong stability is the paper’s stability notion, and Lyapunov drift provides a sufficient condition for establishing it.
- Arrival rates are represented by λ_i, the time-average admissible rate for each node, collected into the vector λ.
- The capacity region Λ is the closed set of arrival-rate vectors characterized by necessary and sufficient stability properties.
- A scheduling algorithm maximizes network throughput when it stabilizes the network for every arrival-rate vector in the interior of Λ.
III. THE TRAFFIC SIGNAL CONTROL PROBLEM
The traffic-signal problem models a road network as links and signalized junctions with movements, phases, and traffic states. At each time slot, the controller selects one phase per junction using locally monitored queues and states to maximize throughput.
- A road network consists of links and signalized junctions, each described by possible movements, phases, and finite traffic states.
- Each phase is a combination of traffic movements that simultaneously receive the right of way.
- Figure 1 lists four example phases for a four-way junction, each pairing incoming and outgoing links into simultaneous movements.
- At the beginning of every slot, the controller chooses the phase p_i(t) for each junction.
- The control objective is maximum network throughput using queue lengths and traffic states supplied by a reliable monitoring system.
IV. BACKPRESSURE-BASED TRAFFIC SIGNAL CONTROLLER
The proposed controller independently selects each junction’s phase using a backpressure-based local computation. The design addresses traffic-specific differences from routing while retaining a distributed implementation and a throughput-oriented guarantee.
- The controller comprises local controllers that are constructed and implemented independently for the junctions.
- The implementation requires synchronized operation on a common clock, despite independent construction and execution across junctions.
- Each local controller selects the phase with the largest computed pressure value among the junction’s feasible phases.
- The phase enumeration is practical because each junction typically has fewer than 10 possible phases.
- Traffic-specific modifications address uncontrolled driver routing, limited phase-based flow authority, and the assumption of infinite link buffers.
V. CONTROLLER PERFORMANCE EVALUATION
The paper characterizes the road network’s capacity region and proves that the proposed backpressure-based controller stabilizes all arrival rates strictly inside it under i.i.d. traffic states.
- Network model: The proof models vehicle movement rates through ξ_i as functions of junction phases, road links, and traffic states.The vector ξ(p,z) collects the rates of all traffic movements under a phase and traffic-state configuration.
- Capacity-region characterization: The capacity region Λ consists of arrival-rate vectors representable by a transmission-rate vector G and flow variables satisfying the network constraints.This characterization is used as the basis for the throughput-optimality proof.
- Stability result: If there exists ϵ > 0 such that λ + ϵ ∈ Λ, the proposed controller stabilizes the network when traffic states are i.i.d. across slots.The proof combines the capacity-region corollary with the controller’s minimization property and a stability proposition.
- Optimality argument: The proposed controller minimizes the final term in the proof’s drift inequality over all alternative policies.This comparison establishes the controller’s key optimization property within the stability argument.
VI. SIMULATION RESULTS
Simulations compare the backpressure-based controller with SCATS across junction, infinite-capacity, finite-capacity, and larger-network settings. The backpressure-based controller reduces queue lengths and supports higher arrival rates under the reported conditions.
- The initial experiment models a 4-phase junction with 4 approaches and 8 links using vehicle-entry data from Singapore loop detectors.
- The simulation implements SCATS and the proposed controller in MATLAB, using SCATS parameters obtained from prior work.
- Under infinite queue capacity, the backpressure-based controller reduces maximum queue length by an order of magnitude compared with SCATS and performs significantly better on average.
- With links limited to 100 vehicles, SCATS supports up to 0.9 times the current arrival rate, whereas the backpressure-based controller supports up to 1.3 times before capacity is exceeded.
- A larger evaluation uses MITSIMLab on a road network with 112 links and 14 signalized junctions, 46 origin-destination pairs, and an arrival rate of 9330 vehicles/hour.
- In the finite-capacity network experiment, the proposed algorithm reduces maximum queue length by a factor of 3 compared with SCATS and performs significantly better on average.Finite link capacity limits queue growth, while queue spillback persists throughout the simulation, especially with SCATS.
VII. CONCLUSIONS AND FUTURE WORK
The paper’s distributed traffic-signal controllers use only local junction information yet formally achieve maximum network throughput. Simulations compare the backpressure-based algorithm with SCATS, while future work addresses fairness, driver response, and coordination.
- Local controllers require no global road-network view or traffic arrival-rate information, yet the algorithm formally achieves maximum network throughput.
- Simulation results show that the backpressure-based algorithm performs significantly better than SCATS.
- Future work includes fairness constraints, mitigating driver-response delays from nonperiodic phase switching, and coordinating signals to produce green waves.