Optimizing Multi-Lane Intersection Performance in Mixed Autonomy Environments

TL;DR

This paper addresses traffic signal control at multilane intersections under mixed HDV/CAV traffic. It proposes a Graph Attention Network–Soft Actor-Critic (GAT–SAC) framework that uses a GAT encoder to capture spatial relations on a traffic graph and a SAC learner to optimize lane-channelization, flow allocation, and phase timing under a multi-objective cost, balancing $D(t)$, $F(t)$, and $S(t)$. Key contributions include explicit modeling of HDV–CAV heterogeneity, fairness-aware rewards, and automated hyperparameter tuning with Tree-structured Parzen Estimators, plus a three-layer integrated control strategy. Empirical results in SUMO show substantial gains: up to 24.1% reduction in average delay and up to 29.2% fewer safety violations compared with traditional baselines, with a non-monotonic yet notable peak performance near moderate CAV penetration around 60%. These findings suggest that GAT–SAC can provide scalable, equitable, and safer intersection management in real-world mixed-autonomy networks and warrant further validation with imperfect sensing and larger networks.

Abstract

One of the main challenges in managing traffic at multilane intersections is ensuring smooth coordination between human-driven vehicles (HDVs) and connected autonomous vehicles (CAVs). This paper presents a novel traffic signal control framework that combines Graph Attention Networks (GAT) with Soft Actor-Critic (SAC) reinforcement learning to address this challenge. GATs are used to model the dynamic graph- structured nature of traffic flow to capture spatial and temporal dependencies between lanes and signal phases. The proposed SAC is a robust off-policy reinforcement learning algorithm that enables adaptive signal control through entropy-optimized decision making. This design allows the system to coordinate the signal timing and vehicle movement simultaneously with objectives focused on minimizing travel time, enhancing performance, ensuring safety, and improving fairness between HDVs and CAVs. The model is evaluated using a SUMO-based simulation of a four-way intersection and incorporating different traffic densities and CAV penetration rates. The experimental results demonstrate the effectiveness of the GAT-SAC approach by achieving a 24.1% reduction in average delay and up to 29.2% fewer traffic violations compared to traditional methods. Additionally, the fairness ratio between HDVs and CAVs improved to 1.59, indicating more equitable treatment across vehicle types. These findings suggest that the GAT-SAC framework holds significant promise for real-world deployment in mixed-autonomy traffic systems.

Figures (7)

• Figure 1: Graph Attention Network–Soft Actor–Critic framework
• Figure 2: Average delay versus CAV penetration rate for different control strategies. The figure compares the average vehicle delay under three control methods—Fixed Timing, MARL-DQN, and GAT-SAC—across varying connected autonomous vehicle (CAV) penetration rates. The GAT-SAC approach consistently achieves the lowest delay, demonstrating improved efficiency as CAV participation increases.
• Figure 3: Safety performance comparison across different control strategies. The figure presents the number of safety violations and normalized violations for Fixed Timing, MARL-DQN, and GAT-SAC control methods. The GAT-SAC approach achieves the lowest safety and normalized violations, indicating enhanced operational safety and stability in mixed-autonomy traffic environments.
• Figure 4: High CAV penetration (60–80%) scenario showing balanced flow and stable queue distributions. Throughput reached 87 vehicles per simulation cycle, indicating efficient coordination under mixed autonomy.
• Figure 5: Moderate CAV penetration (40–50%) scenario dominated by HDVs with visible queuing and less synchronized flow. Throughput achieved 42 vehicles, constrained by limited cooperative control.