Paper Image

Reinforcement learning for following natural language instructions

Published on:

18 February 2023

Primary Category:

Computation and Language

Paper Authors:

Jing-Cheng Pang,

Xin-Yu Yang,

Si-Hang Yang,

Yang Yu

Bullets

Key Details

Proposes developing a specialized task language for reinforcement learning agents to follow instructions

Task language is more compact and consistent than natural language instructions

Enables policies to learn faster, generalize better, integrate with hierarchical RL

Trains a translator to convert natural language to task language

Experiments show benefits over standard approaches that use natural language directly

AI generated summary

Reinforcement learning for following natural language instructions

This paper proposes a new approach for training reinforcement learning agents to follow natural language instructions. It develops a specialized 'task language' that represents instructions in a more compact, consistent way compared to unbounded natural language. This allows policies to learn faster, generalize better, and integrate with hierarchical reinforcement learning. The approach trains a translator to convert natural language to the task language, and a policy to follow the task language. Experiments in a simulated environment demonstrate benefits over standard methods that take in natural language directly.

Answers from this paper

Comments

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

Sign up to comment on this paper

Sign Up