Paper Image

Using language models to guide hierarchical reinforcement learning

Published on:

9 November 2023

Primary Category:

Machine Learning

Paper Authors:

Bharat Prakash,

Tim Oates,

Tinoosh Mohsenin

Bullets

Key Details

Proposes using LLMs like GPT-3 to provide 'common sense' priors for hierarchical RL agents

LLMs guide high-level action selection and exploration for the agent

Evaluated in MiniGrid, SkillHack, Crafter sim environments and on a real robot arm

Outperforms baseline hierarchical RL methods without LLM guidance

Does not require LLM access at deployment time

AI generated summary

Using language models to guide hierarchical reinforcement learning

This paper proposes a method to inject common sense knowledge from large language models into hierarchical reinforcement learning agents. The language models guide high-level action selection, improving exploration and making learning more sample efficient. The approach is tested in simulation and on a real robot arm, outperforming baselines.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up