Problem Motivation

Conventional urban traffic signal control relies on centralized coordination, limited local autonomy, and incomplete real-time awareness. These constraints reduce robustness under demand surges, incidents, sensor noise, and network disruptions, and limit cost-effective scaling across a city.

Goal

DALI-TCS investigates a distributed, autonomous alternative that augments existing traffic infrastructure with software-defined intelligence. The objective is an adaptive and scalable traffic control system that works with standard sensing at intersections (e.g, inductive loops), supports incremental deployment, and remains practical to operate and maintain at city scale.

Core Research Contributions

• Intersection agents: An agent-based control layer that enhances each intersection controller with autonomous decision logic.
• Agent to agent coordination: A communication model that uses available connectivity among controllers to support cooperative behavior.
• Distributed adaptive strategy: A multi-agent control strategy for real-time coordination across intersections, with an emphasis on measurable performance gains, robustness, and scalability.
• Deployment-ready architecture: A system architecture designed for incremental rollout across cities of varying size and infrastructure maturity.

Deployment Evidence

DALI has been deployed in the City of Richardson, where field results show meaningful reductions in delay. Expanded sensing coverage and higher-fidelity sensor data (e.g., from cameras) strengthen the system’s situational awareness and can further improve performance.

System Model

We model an urban traffic network as a graph of signalized intersections connected by roadway links and approaches. Each intersection is controlled by an autonomous software agent that interfaces with the existing signal controller and computes local signal timing decisions in real time. Each agent defines its neighborhood as the set of adjacent intersections connected by road segments with lane-level movement connectivity.

Agents operate using local sensing and data exchanged with neighboring controller agents. Each agent maintains a local representation of intersection state, combining static configuration (intersection geometry and movement definitions), direct observations (signal phase and timing status, vehicle detections), model-derived estimates and predictions (queue lengths and short-horizon arrivals), and model-based inferences (unusual events). Control is distributed: decisions are produced locally at each intersection through coordination across a network of neighboring agents rather than through a central coordinator. The approach supports incremental deployment, since each intersection can be upgraded independently and benefits increase as additional neighboring intersections join the system.

Agent Model

Each DALI TCS agent controls one signalized intersection and interfaces with the existing controller. The agent maintains a local representation of intersection state, produces real time signal timing decisions, and coordinates with neighboring agents through direct communication.

The agent state includes:

• Signal state: active phase, timing status, and controller timing parameters that constrain feasible operation.
• Traffic state: vehicle detections, queue length estimates, and short horizon arrival estimates for relevant approaches or movements.
• Coordination state: recent neighbor messages, inferred upstream releases, and local neighborhood context used for coordination.

The agent’s perception model includes:

• Static configuration: intersection geometry, lane configuration, and movement definitions.
• Direct observations: controller phase and timing status plus baseline sensor events.
• Model derived signals: queue estimates, short horizon arrival predictions, and unusual event inferences derived from data analysis and trained models.

The agent’s actions differ by traffic regime:

• Normal traffic conditions: selection of a feasible near term timing plan over a short horizon, subject to controller and safety constraints.
• Congested traffic conditions: proposal of split adjustments and participation in neighborhood coordination prior to applying changes.

The agent’s goals emphasize:

• Performance: measurable reductions in delay and queue growth.
• Robustness: stable behavior under demand surges, sensing noise, and network disruptions.
• Scalability: effective operation as deployment expands across intersections.
• Operational feasibility: strict compliance with controller timing and safety constraints.

Communication & Coordination Model

DALI TCS uses direct agent to agent communication among neighboring intersection controllers. Communication is peer to peer and does not rely on a central coordinator.

Neighborhood scope. Communication occurs among:

• Adjacent intersection agents defined by road segment connectivity and lane level movement connectivity.

Information exchanged. Messages include traffic summaries that support coordination, such as:

• Upstream detection timing and movement level flow information.
• Turning information when available, used to estimate downstream arrivals.

Role of coordination. Communication supports:

• Arrival estimation over a short horizon to improve local decisions.
• Cooperative timing behavior across adjacent intersections.
• Congestion coordination during split adjustment and agreement processes.

Communication timing. Exchanges occur in real time and align with control needs, such as:

• Detection driven updatesand control cycle updates, depending on deployment configuration.

Decision Making & Control Strategy

DALI TCS uses a distributed planning and control loop at each intersection. Each agent generates candidate timing plans for the near future, evaluates them using local and neighbor informed estimates, executes the best plan, and repeats the process.

Planning horizon. Decision making is performed over:

• An observable horizon that reflects the portion of upstream traffic that can influence near term decisions at the intersection.

Timing plan representation. A plan is defined as:

• A sequence of feasible phase combinations, where each plan element specifies start time, end time, and green duration.
• Plans that satisfy controller and safety constraints, including minimum and maximum green and clearance intervals.

Plan evaluation. Each candidate plan is scored using:

• Queue related terms derived from queue length estimates.
• Arrival related terms derived from short horizon arrival estimates.

Real time execution. The agent performs:

• Selection of the best scoring plan, followed by execution through the controller interface.
• Replanning in a receding horizon manner.

Congestion mode. Under congestion, the agent shifts to coordinated split adjustment:

• The initiating agent derives a split adjustment proposal intended to relieve congestion.
• Neighboring agents evaluate local impact and provide feedback.
• The coordination process propagates when congestion impact extends beyond immediate neighbors.
• The initiating agent decides to apply or ignore the configuration and informs affected agents.

Assumptions & Constraints

Sensing. DALI TCS assumes:

• Baseline detection such as inductive loops at intersections.
• Queue estimation quality that depends on detector placement and controller state observability.
• Improved state fidelity with richer sensing sources, such as camera-based sensing, when available.

Communication. The approach assumes:

• Real time message exchange among adjacent controllers.
• No requirement for global state collection or centralized decision making.

Controller and safety constraints. Decisions must:

• Remain feasible under controller timing limits and safety requirements, including clearance intervals.

Real time operation. Computation must:

• Fit within operational control time scales at the intersection.

Infrastructure and deployment. The architecture supports:

Incremental rollout where intersections can be upgraded independently and coordination benefits increase as neighboring intersections join.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.

Behnam Torabi, Rym Z. Wenkstern, and Robert Saylor. A collaborative agent-based traffic signal system for highly dynamic traffic conditions. Journal of Autonomous Agents and Multi-Agent Systems (JAAMAS), 34(1):1–24, 2020.

Behnam Torabi, Rym Z. Wenkstern, and Robert Saylor. A collaborative agent-based traffic signal system for highly dynamic traffic conditions. Journal of Autonomous Agents and Multi-Agent Systems (JAAMAS), 34(1):1–24, 2020.

Behnam Torabi, Rym Z. Wenkstern, and Robert Saylor. A collaborative agent-based traffic signal system for highly dynamic traffic conditions. Journal of Autonomous Agents and Multi-Agent Systems (JAAMAS), 34(1):1–24, 2020.

Behnam Torabi, Rym Z. Wenkstern, and Robert Saylor. A collaborative agent-based traffic signal system for highly dynamic traffic conditions. Journal of Autonomous Agents and Multi-Agent Systems (JAAMAS), 34(1):1–24, 2020.