Three Phase Traffic Negotiation Framework for Autonomous and Human-Driven Environments in Heterogeneous Smart Cities

Managing traffic in mixed-autonomy environments is challenging due to the coexistence of human-driven vehicles, connected and autonomous vehicles (CAVs), and communication uncertainties. This study proposes a three-phase traffic control framework that integrates Dynamic Control Zone (DCZ) selection, which adaptively designates lead CAVs for local coordination using queue lengths, approach speeds, and signal timings; Model Predictive Control (MPC), which predicts short-term traffic dynamics and enforces safety and operational constraints; and Deep Reinforcement Learning (DRL), which learns long-term adaptive strategies to improve efficiency, environmental impact, and stability. Furthermore, to enhance adaptability, DCZ parameters are tuned using Self Adaptive Ninja Optimizer (SANO) for multi-objective performance. The framework is implemented in MATLAB/Simulink on simulated networks inspired by real-world traffic data, incorporating V2X imperfections to reflect communication conditions. The proposed framework highlights the novelty of combining MPC’s short-term safety, DRL’s long-term adaptability, and DCZ’s adaptive selection into a unified traffic management framework, providing a promising step toward mixed-autonomy traffic control in next-generation cities.