Text-Driven Simulation Scene Generation

Overview

This project focuses on end-to-end simulation scene generation driven by natural language, enabling users to create complex 3D environments directly from textual descriptions.

Role

• Technical Lead

Funding Source

• Guangdong Flagship Project

Contributions

• Architected the full pipeline for text-driven scene generation, from language understanding to 3D scene synthesis.
• Integrated Unreal Engine (UE) with 2D Gaussian Splatting, automating the construction of high-fidelity 3D scenes from natural language descriptions.
• Designed reusable modules for scene layout, asset placement, and lighting control, making the system extensible to new environments and tasks.

Techniques & Tools

• Natural Language Processing for scene and object specification
• Unreal Engine (UE) for high-fidelity simulation
• 2D Gaussian Splatting for fast, realistic rendering of complex scenes