What happened
Researchers developed a two-stage method where a robot first learns a policy (a decision-making pattern) from examples, then uses that learned pattern to guide a world model planner (a system that predicts what will happen next) toward goals specified in natural language. This means robots can navigate complex environments by combining intuition from learned behavior with forward-looking planning, rather than relying on either alone.
Why it matters
This is an incremental advance in embodied AI — robots that can follow instructions in the real world — but it does not demonstrate that the approach works at meaningful scale, solves a previously unsolved problem, or changes what's deployable. It's a laboratory result showing one combination of existing techniques outperforms baselines on a research dataset.