VLALight: Lightweight Vision-Language-Action Models for Emergency-Aware Traffic Signal Control

2026-09-24

AI Summary

This study addresses the loss of visual details and high inference latency in traffic signal control, which hinder real-time responses to emergency vehicles. To this end, it proposes a lightweight, end-to-end Vision-Language-Action (VLA) framework. By eliminating intermediate textual descriptions and handcrafted features, the method integrates intersection observations and phase information through unified multi-view inputs, leveraging a 0.5B-parameter large language model to directly map visual data to discrete signal actions. Experimental results demonstrate that the proposed framework achieves real-time inference on local hardware, reduces emergency vehicle waiting time by 21.1% compared to cascaded approaches, and exhibits strong generalization capabilities to unseen intersection topologies and traffic flow patterns.

Abstract

Traffic signal control (TSC) is essential for mitigating urban congestion. Recent advances in vision-language models (VLMs) enable richer interpretation of intersection scenes, opening new opportunities for visual-context-aware TSC. However, the loose coupling and repeated information conversion between modules can lead to the loss of fine-grained visual details, while sequential inference introduces substantial latency. To address these limitations, we propose VLALight, a lightweight end-to-end vision-language-action framework that directly maps intersection observations and signal-phase information to discrete signal actions. To handle the multi-view nature of TSC, VLALight combines multiple directional camera views into a unified visual input and uses textual instructions to establish their correspondence with traffic movements and signal phases. This design enables direct action prediction with a compact 0.5 B-parameter model, without intermediate image-to-text descriptions or handcrafted traffic-state representations. Experiments show that VLALight delivers the best emergency-vehicle service of all compared methods, reducing pooled emergency waiting time by 21.1% over the cascaded VLMLight while running in real time on local hardware and generalizing to unseen intersection topologies and traffic-flow patterns.

Problem

• Traffic Signal Control
• Vision-Language Models
• Emergency Vehicle
• Inference Latency
• Multi-view

Innovation

• Vision-Language-Action Models
• Traffic Signal Control
• End-to-End Framework
• Lightweight Model
• Emergency Vehicle Priority