TL;DR
This paper proposes a novel linear programming model and a rolling optimization algorithm to schedule connected automated vehicles (CAVs) at unsignalized intersections, aiming to minimize vehicle delays and avoid collisions. The approach significantly reduces average vehicle delay compared to traditional methods and offers a faster solution time than commercial solvers, making it practical for real-world deployment.
What Problem Does This Paper Address?
Intersections are notorious bottlenecks in urban traffic, contributing to congestion and accidents. While traffic signals are a common solution, they can be inefficient. Connected automated vehicles (CAVs) offer a promising avenue for optimizing intersection management by enabling intelligent coordination. However, managing CAVs at unsignalized intersections presents challenges: ensuring safety by preventing conflicts between vehicles from different directions and improving traffic flow efficiency. Existing methods for unsignalized intersections often struggle with computational complexity, especially for longer time horizons, and lack consideration for the dynamic nature of traffic. This paper tackles the problem of optimizing the sequence and timing of CAVs at an unsignalized intersection to minimize delays while guaranteeing safety.
Methods
The core of this work is a linear programming (LP) model designed to optimize the sequence of vehicles passing through the intersection. The objective function minimizes the total vehicle delay within a defined optimization time window. Key constraints include:
- Arrival Time Bounds: Ensuring vehicles arrive within a reasonable timeframe based on free-flow speed.
- Minimum Headway: Maintaining a safe time gap between successive vehicles from the same direction.
- Conflict Zone Safety: Preventing vehicles from conflicting directions from entering the intersection's conflict zone simultaneously, enforced through a minimum time difference.
To address the computational burden of solving this LP model for large numbers of vehicles or long time windows, the authors introduce a rolling optimization algorithm. This approach breaks down the problem into smaller, sequential optimization tasks over shorter time windows. This "moving horizon" strategy allows the algorithm to adapt to changing traffic conditions dynamically. The algorithm ensures continuity between time windows by enforcing safety constraints between the last vehicle of a previous window and the first vehicle of the next.
Datasets
Not discussed in the paper. The study relies on simulated traffic scenarios.
Key Insights and Ideas
- LP for Intersection Scheduling: Framing the unsignalized intersection scheduling problem as an LP problem is a structured way to balance safety and efficiency.
- Rolling Optimization for Scalability: Decomposing the problem into smaller, sequential optimization windows is crucial for making the solution computationally feasible for real-time applications. This is a common strategy in control theory but is effectively applied here to traffic management.
- Trade-off between Delay and Computation Time: The study highlights that there's an optimal time window length that balances minimizing vehicle delay and keeping computation time manageable.
- Importance of Headway Parameters: The sensitivity analysis reveals that the difference between safe headway times for same-direction and conflicting-direction vehicles significantly impacts the effectiveness of the proposed model. A larger difference leads to better delay reduction.
Key Results
- Significant Delay Reduction: The proposed model and algorithm achieved substantial reductions in average vehicle delay, with a maximum reduction of 76.22% compared to the First-Come-First-Served (FCFS) method in high-demand scenarios.
- Competitive Performance with Solvers: The proposed algorithm's performance in terms of delay reduction was comparable to the commercial Gurobi solver.
- Faster Solution Times: Crucially, the rolling optimization algorithm significantly reduced solution times compared to Gurobi, especially under high traffic demand, making it more practical for real-time control. For instance, in scenario 3, Gurobi took over 43,200 seconds, while the proposed algorithm took 415.38 seconds.
- Optimal Time Window Identified: The study found an optimal optimized time window length of 20 seconds that balances average vehicle delay and computation time.
Limitations and Future Work
- Simplified Intersection Model: The current model assumes a simple four-way intersection and does not consider turning vehicles, which adds complexity to real-world scenarios.
- Limited to Isolated Intersections: The model is designed for isolated intersections, and its applicability to networks of closely spaced intersections needs further investigation.
- Future Work: The authors suggest extending the model to handle more complex intersections (e.g., multi-turn intersections) and developing algorithms for optimizing traffic flow across multiple interconnected intersections or intersection groups. They also acknowledge that for very high traffic demands, the solution time might still exceed practical limits, suggesting further optimization of the algorithm or time window length.
Summary updated