💻 computer science

Controlling Traffic without Tolls: A Non-Monetary Framework for Autonomous Intersections

This paper proposes a non-monetary, infrastructure-light framework for autonomous intersection management that uses timestamp-based scheduling to steer traffic toward socially efficient flows, achieving a 71% reduction in the efficiency gap between user and system optima on the Sioux Falls network by establishing a unique equilibrium through a congestion-game formulation.

Original authors: Arda Kosay, Yusuf Saltan, Jyun-Jhe Wang, Chung-Wei Lin, Muhammed O. Sayin

Published 2026-04-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Arda Kosay, Yusuf Saltan, Jyun-Jhe Wang, Chung-Wei Lin, Muhammed O. Sayin

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a busy city where every driver is trying to get to work as fast as possible. Usually, they all pick the route they think is shortest. But when everyone does this, they all end up stuck in the same traffic jam. This is the classic "tragedy of the commons" in traffic: individual logic leads to a collective disaster.

Traditionally, cities try to fix this with tolls (charging money to use a road) or by building more roads. But tolls feel unfair to poorer drivers, and building roads is expensive and slow.

This paper proposes a clever, free, and non-monetary solution using the technology of the future: Autonomous Vehicles and Smart Intersections.

Here is the simple breakdown of how it works, using some everyday analogies.

1. The Problem: The "Selfish Driver"

Think of traffic like a game of musical chairs, but the chairs are empty seats on a highway. Every driver wants the best seat (the fastest route). If everyone rushes for the "fastest" looking route, it gets crowded, and suddenly it's the slowest route.

2. The New Tool: The "Smart Traffic Cop" (The RSU)

In this future, every intersection has a Roadside Unit (RSU). Think of the RSU as a super-smart traffic cop that talks directly to every car.

Old Way: The cop just lets cars through in the order they arrive (First-Come, First-Served).
New Way: The cop can subtly tweak the "arrival time" of a car before letting it through.

3. The Secret Sauce: "Time Travel" for Cars

The core idea is Timestamp Modification.
Imagine you are at a coffee shop. You order a latte. The barista (the RSU) usually serves you based on who ordered first.

The Trick: The system can tell the barista, "Hey, pretend this person ordered 10 seconds later than they actually did," or "Pretend they ordered 10 seconds earlier."
The Result: If you pretend to order later, you have to wait a bit longer for your coffee. If you pretend to order earlier, you get served sooner.

In the paper's system:

Negative Incentive (The "Wait"): If a car is trying to take a route that is already too crowded, the RSU tells the car, "You're actually 10 seconds late." The car has to wait a bit longer at the intersection. This makes that route feel "expensive" in terms of time.
Positive Incentive (The "Fast Pass"): If a car chooses a less crowded, alternative route, the RSU says, "You're actually 10 seconds early!" The car gets through the intersection faster.

4. The Strategy: The "Grandmaster" vs. The "Players"

The paper uses a Two-Layer Strategy (Bilevel Optimization):

Layer 1 (The Players): The drivers (or their self-driving software) are selfish. They just want the fastest trip. They see the "wait times" created by the RSUs and naturally switch to the faster routes.
Layer 2 (The Grandmaster): A central computer (the System Planner) watches the whole city. It doesn't force anyone to go where they don't want to. Instead, it plays a game of chess.
- It calculates: "If I make Route A feel 10 seconds slower, will enough drivers switch to Route B to clear the jam?"
- It then subtly adjusts the "arrival times" (timestamps) at the intersections to nudge drivers toward the perfect balance where the whole city moves fastest, even if no single driver is getting the absolute fastest trip possible.

5. Why This is a Big Deal

No Money Involved: You don't need to charge drivers. You just use time.
No New Roads: You don't need to pave new asphalt. You just use the existing smart infrastructure better.
Fairness: It doesn't punish the poor; it just nudges everyone toward a better flow.
The Result: In their computer simulations (using a famous test network called "Sioux Falls"), they showed that this method could close 71% of the gap between "chaotic traffic" and "perfectly organized traffic."

The Bottom Line

Think of this system as a conductor for a traffic orchestra.
Instead of shouting at the musicians (drivers) to play louder or softer, the conductor (the RSU) slightly adjusts the tempo for specific instruments. By the time the musicians hear the change, they naturally adjust their playing to fit the song perfectly. The result? A symphony of smooth traffic flow, achieved without spending a single dollar or building a single new road.

1. Problem Statement

Urban transportation systems are becoming increasingly complex due to the integration of Connected and Automated Vehicles (CAVs) and autonomous intersections. While autonomous intersections allow for fine-grained, vehicle-level scheduling via Roadside Units (RSUs), they introduce a strategic challenge:

The Efficiency Gap: Self-interested drivers choose routes to minimize their own travel time (User Equilibrium - UE), which often leads to suboptimal network-wide performance compared to the System Optimal (SO).
The Limitation of Current Solutions: Traditional methods to close this gap rely on monetary incentives (tolls), which raise equity concerns, or infrastructure expansion, which is costly and slow.
The Research Question: How can intersection-level control be systematically designed to influence routing behavior and improve network-wide efficiency without relying on monetary incentives or new physical infrastructure?

2. Methodology

The authors propose a hierarchical, non-monetary control framework that leverages timestamp modifications at autonomous intersections to act as "virtual tolls."

A. System Architecture

The framework separates control into two layers (Fig. 2):

Low-Level (Local): RSUs at individual intersections manage vehicle passage using standard policies (e.g., First-Come-First-Serve or optimal control) based on vehicle timestamps.
High-Level (System Planner): A central planner modifies the timestamps of vehicle requests before they reach the RSU.
- Mechanism: The planner can artificially delay (negative incentive) or advance (positive incentive) a vehicle's timestamp.
- Effect: A delayed timestamp causes the RSU to schedule the vehicle later, increasing its expected waiting time at that intersection. This acts as a path-dependent cost, steering drivers away from congested routes.

B. Analytical Framework

Modeling: The traffic network is modeled as a directed graph where nodes represent autonomous intersections and edges represent roads.
Cost Function: The cost of a path $p$ $p$ is the sum of edge costs (travel time) and node costs (intersection delay).
- The node cost is defined as $c_v^p(f_v, u_v^p) = c_v(f_v) + u_v^p$ , where $u_v^p$ is the additive timestamp modification (incentive).
Equilibrium: Drivers act selfishly to minimize their perceived cost, leading to a Wardrop Equilibrium.
Key Theoretical Insight: The authors prove that timestamp modifications act as additive, path-specific costs. This allows the system to be modeled as a congestion game with a potential function.

C. Optimization Formulation

The problem is formulated as a Bilevel Optimization problem:

Lower Level (Drivers): Minimize the potential function $\Phi(f, u)$ to find the User Equilibrium flow $f(u)$ for a given incentive profile $u$ .
Upper Level (Planner): Select the incentive profile $u$ (subject to bounds $[-\tau, \tau]$ ) to minimize the total Social Cost $C(f, u)$ .

Crucial Theoretical Contribution: The authors establish the existence and essential uniqueness of the equilibrium flow. This means that while multiple flow distributions might mathematically satisfy equilibrium conditions, they all yield the same social cost. This eliminates ambiguity in the bilevel optimization, making the problem tractable and scalable.

3. Key Contributions

Scalable Hierarchical Architecture: A novel separation of local intersection scheduling from network-level coordination, avoiding the computational intractability of end-to-end centralized control.
Non-Monetary Timestamp Mechanism: A method to induce additive cost adjustments along routes using existing RSU capabilities, avoiding the need for tolls or new hardware.
Theoretical Guarantees: Proof of the essential uniqueness of equilibrium flows under this specific cost structure, enabling a well-defined bilevel optimization framework.
Experimental Validation: Demonstration of the framework's effectiveness using high-fidelity microscopic simulations.

4. Experimental Results

The framework was tested on two networks: the theoretical Braess Network and the realistic Sioux Falls Network (24 nodes, 76 links, 552 OD pairs).

Simulation Setup:
- Used the SUMO (Simulation of Urban MObility) platform to model vehicle dynamics and intersection queuing.
- Calibrated intersection delays using quartic cost functions fitted to simulation data (matching the BPR edge cost model).
- Validated that timestamp modifications act as additive offsets to node costs (verified via parallel cost curves in Fig. 7).
Performance Metrics (Sioux Falls):
- Baseline: The User Equilibrium (without incentives) had a social cost of $8.04 \times 10^6$ min.
- System Optimal: The theoretical minimum cost was $7.75 \times 10^6$ min.
- High-Intervention Case ( $u \in [0, 2]$ min): The incentivized equilibrium reached $7.84 \times 10^6$ min, closing 71.1% of the efficiency gap between UE and SO.
- Low-Intervention Case ( $u \in [0, 0.5]$ min): Even with strict limits (max 30-second delays), the system closed 51.3% of the gap.
Flow Redistribution: Visualizations (Fig. 9) showed that incentives successfully diverted traffic from over-saturated central corridors to peripheral routes, balancing network load without physical changes.

5. Significance and Future Directions

Practicality: This approach requires no new infrastructure (only software updates to RSUs) and no monetary transactions, addressing equity and deployment barriers associated with tolls.
Scalability: The "essential uniqueness" of the equilibrium ensures that the optimization problem remains solvable even in large networks, a significant improvement over previous game-theoretic traffic models.
Future Work: The authors suggest extending the framework to mixed-autonomy environments (human + autonomous vehicles), integrating real-time adaptive learning (Reinforcement Learning), and balancing efficiency with fairness constraints.

In conclusion, the paper demonstrates that strategic manipulation of intersection scheduling timestamps is a powerful, scalable, and equitable tool for optimizing urban traffic flow in the era of autonomous vehicles.