Contrastive Learning from Exploratory Actions: Leveraging Natural Interactions for Preference Elicitation

Nathaniel Dennler, Stefanos Nikolaidis, Maja Matarić · 2025

Read Contrastive Learning from Exploratory Actions: Leveraging Natural Interactions for Preference Elicitation by Nathaniel Dennler, Stefanos Nikolaidis, Maja Matarić (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-2501-01367.txt.

Description

Teaching a robot your taste usually means tedious labeling, yet people who explore a robot's behaviors show what they like and what they pass by. Nathaniel Dennler, Stefanos Nikolaidis, and Maja Matarić in 2025 turn that exploration into training data. Their method, CLEA, sets the ones a user lingers on against those the user skips, and learns features that match what users care about.

Features learned from twenty-five people's free search with a Kuri robot were tested on forty-two fresh users. They beat self-supervised features on completeness, simplicity, minimality, and explainability when drawing out preferences. Used this way, that search cuts the time it takes to teach a robot to suit one user.

Catalog record

CREATORNathaniel Dennler, Stefanos Nikolaidis, Maja Matarić
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE68 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-01367

Browse the full library · About the Tower of Babel