Introduction
Coordinating Connected and Automated Vehicles (CAVs) at conflict zones—such as non-signalized intersections or highway merges—is a critical challenge for future autonomous infrastructure. While many existing strategies prohibit lane changes to simplify the coordination problem, real-world scenarios like road work zones make lateral maneuvers mandatory. Traditional planning-based solutions, often formulated as Mixed-Integer Linear Programming (MILP), are computationally expensive and struggle to scale in real-time as vehicle density increases. This paper proposes a novel bi-level coordination strategy designed to handle flexible lane changes efficiently while maintaining safety and high traffic throughput.
The Bi-Level Planning Framework
To address the computational complexity of simultaneous longitudinal and lateral coordination, the authors decouple the problem into two distinct levels:
1. Upper-Level: Passing Order Optimization
The upper level determines the "right-of-way" by establishing a passing sequence for all vehicles within the coordination zone. Instead of an exhaustive search, which would be computationally prohibitive, the framework utilizes Monte Carlo Tree Search (MCTS). By representing the solution space as a tree and applying heuristic rules—such as prioritizing leading vehicles—MCTS can quickly identify high-quality passing orders within a 0.5-second planning window.
2. Lower-Level: Trajectory Generation
Once a passing order is established, the lower level calculates conflict-free trajectories. It employs a "passing-order-to-trajectory interpretation algorithm" that selects from a library of smooth lane-change trajectories modeled using 5th-order polynomials. To ensure safety during these maneuvers, the road is discretized into a cell-based model. A lane-changing vehicle is assumed to occupy cells in both its origin and destination lanes simultaneously until the maneuver is complete.
A key innovation in this framework is the classification of conflict zones. As shown in the figure above, critical conflict zones refer to fixed bottlenecks like intersections or lane closures, while local conflict zones are temporary risks created by specific vehicle maneuvers, such as lane changes. This distinction allows the bi-level approach to manage global traffic flow while ensuring local safety.
Simulation and Performance Evaluation
The strategy was evaluated through numerical simulations implemented in C++. Vehicle arrivals were modeled as a Poisson process to simulate varying traffic demands, ranging from 360 to 2400 vehicles per hour. The primary test scenario was a three-lane highway work zone where one lane was closed, forcing vehicles to merge into the remaining lanes.
Results and Discussion
The proposed bi-level strategy demonstrated significant improvements over the standard First-In-First-Out (FIFO) baseline:
- Significant Delay Reduction: At peak traffic volumes (2400 veh/h), the average delay per vehicle was reduced from 6.89 seconds to 3.34 seconds—a performance gain of over 50%.
- Increased Throughput: The strategy successfully processed more vehicles in high-density scenarios (382 vs. 372 in a 10-minute window) and produced smoother velocity profiles, contributing to better fuel economy and passenger comfort.
- Real-Time Feasibility: By searching approximately 200 nodes within 0.5 seconds, the algorithm proved capable of real-time implementation, meeting the strict timing requirements of autonomous traffic coordination.
Conclusion and Future Work
This research provides a scalable solution for CAV coordination in complex environments where lane changes are necessary. By combining MCTS with a fast trajectory interpretation algorithm, the system achieves near-optimal flow without the computational burden of traditional MILP methods.
Future work will focus on extending the model to handle mixed traffic environments (including human-driven vehicles) and testing the strategy on more complex multi-modal intersection geometries. Additionally, the authors are developing CAV prototypes for real-world field validation.
Summary updated