SimWorld Studio: AI Platform Generates 3D Environments for Embodied Agents
SimWorld Studio, an open-source initiative utilizing Unreal Engine 5, has launched a feature for the automatic creation of dynamic 3D environments designed for embodied agent learning. At its heart lies SimCoder, a coding agent enhanced with tools and skills that generates and executes engine-level code to build realistic worlds based on language or image prompts. This agent evolves autonomously by learning from verifier feedback, such as compilation errors. The platform seeks to fill the gap in diverse, automatically generated 3D environments for interactive learning, as current simulators depend on manually designed scenes or procedural templates. Unlike recent LLM-based systems that yield static scenes, SimWorld Studio aspires to offer immersive training environments similar to sandboxes for advanced digital agents. This project is outlined in arXiv paper 2605.09423.
Key facts
- SimWorld Studio is an open-source platform built on Unreal Engine 5.
- It generates evolving embodied learning environments automatically.
- SimCoder is a tool/skill-augmented coding agent that writes and executes engine-level code.
- SimCoder constructs 3D worlds from language or image instructions.
- SimCoder self-evolves using verifier feedback like compilation errors.
- Existing embodied simulators rely on manually crafted scenes or procedural templates.
- Recent LLM-based 3D generation systems mainly produce static scenes.
- The platform aims to provide diverse, automatically generated 3D environments for embodied agents.
Entities
Institutions
- arXiv