We propose a unified framework that integrates 5G and emerging 6 G networks with machine learning (ML) and multi-access edge computing (MEC) for autonomous electric vehicle (AEV) systems. Designed to meet ultra-reliable lowlatency communication (URLLC) demands, the architecture comprises four distinct layers Vehicle, Communication, Edge Computing, and Application to support intelligent and scalable V2X communication in IoT-based vehicular environments. The system deploys advanced ML models, including LSTM, CNN, Deep Q-Networks, Support Vector Regression, and Gradient Boosting, on GPU-enabled MEC nodes to enable real-time decision-making, anomaly detection, and energy-aware Vehicle-to-Grid (V2G) operations. We evaluate the framework using a co-simulation environment based on NS-3 and SUMO, simulating 200 AEVs within a dynamic 5km×5km urban grid. Results demonstrate up to 22.5 % energy savings, latency under 5 ms, a packet delivery ratio of 98.3 %, and anomaly detection accuracy of 97.2 %, significantly o utperforming c loud-based a nd localonly baselines. The proposed architecture lays the foundation for secure, efficient, a nd s calable n ext-generation intelligent transportation systems.