Text this: LLMs augmented hierarchical reinforcement learning with action primitives for long-horizon manipulation tasks.