AI-Enabled Multi-Scale Simulation Framework for Autonomous Transportation System: Cross-Layer Digital Twins and Modular Bus Applications

Autonomous Transportation Systems (ATS) require simulation technologies capable of spanning traffic-system-level dynamics, vehicle-level motions, and component-level mechanical responses. However, traditional transportation engineering simulations and vehicle engineering simulations operate at disconnected temporal-spatial scales, making cross-scale reasoning and system-level performance optimization extremely challenging. This paper proposes an AI-enabled (Artificial Intelligence), multi-module, multi-scale simulation framework for ATS. The framework unifies macro-level traffic flow modeling, meso-level interaction analysis, and micro-level vehicle dynamics using a continuous cross-scale digital-twin architecture. We design high-throughput sensor-data ingestion, real-time co-simulation engines, and a scalable model-switching algorithm capable of smooth transitions among system-, vehicle-, and component-level simulations. In addition, we develop multi-engine cooperative computing strategies to coordinate traditional physics-based simulation with AI-driven prediction and resource optimization. A representative application-autonomous modular buses capable of dynamic in-motion connection and separation-is presented to demonstrate the proposed framework.