When Digital Twins Meet Large Language Models: Realistic, Interactive, and Editable Simulation for Autonomous Driving

Abstract: Simulation frameworks have been key enablers for the development and validation of autonomous driving systems. However, existing methods struggle to comprehensively address the autonomy-oriented requirements of balancing: (i) dynamical fidelity, (ii) photorealistic rendering, (iii) context-relevant scenario orchestration, and (iv) real-time performance. To address these limitations, we present a unified framework for creating and curating high-fidelity digital twins to accelerate advancements in autonomous driving research. Our framework leverages a mix of physics-based and data-driven techniques for developing and simulating digital twins of autonomous vehicles and their operating environments. It is capable of reconstructing real-world scenes and assets (real2sim) with geometric and photorealistic accuracy and infusing them with various physical properties to enable real-time dynamical simulation of the ensuing driving scenarios. Additionally, it also incorporates a LLM interface to flexibly edit the driving scenarios online via natural language prompts. We analyze the presented framework in terms of its fidelity, performance, and serviceability. Results indicate that our framework can reconstruct 3D scenes and assets with up to 97% structural similarity, while maintaining frame rates above 60 Hz. We also demonstrate that it can handle natural language prompts to generate diverse driving scenarios with up to 95% repeatability and 85% generalizability.