Urban traffic signal management systems often rely on outdated techniques and strategies that fail to adapt to dynamic roadway conditions, leading to safety concerns, congestion, and access issues for road users. In addition, current signal optimization approaches rarely consider energy efficiency as the main objective. This research proposes a next-generation safe, efficient and reliable traffic signal control framework powered by advanced roadside sensing and foundation models, specifically Visual Language Models (VLMs) and Multi-Modal Large Language Models (MMLLMs). By integrating high-definition cameras, LiDAR, and real-time data analytics, the system will accurately detect multimodal traffic flows, predict future traffic conditions, and optimize signal phase and timings to enhance mobility while minimizing energy consumption. The framework will be validated through a case study at the Riverside Smart Intersection testbed, leveraging real-world data and co-simulation environments.