Fig. 1 | Scientific Reports

Fig. 1

From: LLMs augmented hierarchical reinforcement learning with action primitives for long-horizon manipulation tasks

Fig. 1

Framework of LARAP that enables the robot to leverage action primitives to solve manipulation tasks with LLMs guidance effectively and efficiently. The LLMs are used to guide the high-level policy and accelerates learning. It is prompted with the context, skill informance, and the current task and observation. The LLM’s output biases high-level action selection.

Back to article page