What happened
Large language models (LLMs) usually struggle to understand the physical world when controlling robots using only text. A new method teaches these AI models to use programming concepts and diagrams instead of just natural language. This means robots can plan complex tasks more reliably and execute them better than before.
Why it matters
Robots need to understand how objects relate in space and how actions cause changes. Text-based AI often misses these details, leading to clumsy or failed tasks. This method gives AI a structured way to map the world, much like a human engineer would, which could make robots much more capable in complex, real-world environments.