Learning Parameterized Skills from Demonstrations

Vedant Gupta, Haotian Fu, Calvin Luo, Yiding Jiang, George Konidaris · 2025

Read Learning Parameterized Skills from Demonstrations by Vedant Gupta, Haotian Fu, Calvin Luo, Yiding Jiang, George Konidaris (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-24095.txt.

Description

Vedant Gupta and four co-authors present DEPS, a method that learns reusable skills from expert demonstrations. Instead of copying single actions, the system learns skills that take parameters, so one skill can be tuned for many objects or goals. A higher level policy picks which skill to use and what values to give it at each step.

The method learns the skills and the picker together, using tools that keep the skills distinct, long lasting, and meaningful. On the LIBERO and MetaWorld test suites, DEPS beat both multitask and other skill learning methods on new tasks. It also found skills a person can read clearly, such as grasping an object in a set way.

Catalog record

CREATORVedant Gupta, Haotian Fu, Calvin Luo, Yiding Jiang, George Konidaris
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE83 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-24095

Browse the full library · About the Tower of Babel