A method for constructing an optimal behavior basis for the Option Keyboard, enabling zero-shot identification of optimal solutions for any linear-reward task.

We use LLMs to create state-of-the-art AI planners.