Towards Simulation-Ready LLM-Generated Articulated Scenes 迈向仿真就绪的 LLM 生成关节场景

Training robot manipulation policies needs large, diverse collections of simulation-ready articulated assets, which hand-curated datasets cannot supply at scale. This project builds an LLM-agent workflow that generates such scenes from natural language.

训练机器人操作策略需要大量、多样的仿真就绪关节资产，而人工标注的数据集难以规模化提供。本项目构建了一套 LLM 智能体流程，从自然语言生成这类场景。

Code not yet public.

代码暂未公开。