Vehicular ad hoc networks (VANETs) demand seamless integration of mobility
modeling, network optimization, and edge intelligence to meet the requirements of
next-generation intelligent transportation systems. Existing studies often treat these
domains separately, limiting realism and deployment readiness. This paper presents a
co-simulation framework for network-optimized ad hoc communication using an
inverted AI edge simulation approach. The framework in CARLA driving simulator
with OMNeT++is to synchronize vehicle dynamics and network behavior, while a
lightweight YOLOv8-Nano model deployed at the edge enables real-time object
detection under constrained resources. Transformer-based architectures are
incorporated for collaborative V2X perception, enhancing resilience against
occlusion and dense traffic conditions. To ensure reliable connectivity, adaptive
multi-network communication strategies are applied, integrating IEEE 802.11p, LTE-
V2X, and 5G-V2X in compliance with ETSI and Third Generation Partnership
Project (3GPP) standards. Experimental results demonstrate reductions in latency,
improvements in detection accuracy, and higher packet delivery reliability compared
to traditional approaches. The proposed framework establishes a scalable foundation
for deploying edge-enhanced, AI-driven vehicular networks through realistic co-
simulation.