Paper Image

Language-guided robot skill learning

Published on:

26 July 2023

Primary Category:


Paper Authors:

Huy Ha,

Pete Florence,

Shuran Song


Key Details

Uses language guidance for efficient robot exploration

Decomposes tasks recursively into motion primitives

Verifies and retries failed executions

Distills exploration data into reusable policies

Improves success rate over data collection policy

AI generated summary

Language-guided robot skill learning

This paper presents a framework that uses language guidance from large language models to efficiently generate labeled robot data, then distills it into reusable visuomotor policies. The language model decomposes tasks, grounds them into robot motion primitives, and verifies success. This enables autonomous exploration and labeling. The distilled policy takes visual observations and language descriptions as input, and outputs control sequences, improving on the data collection policy's performance.

Answers from this paper


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

Sign up to comment on this paper

Sign Up