Sangoh Lee (@sangoh64166) / X

Pinned

Dec 24, 2025

Imagine asking for your favorite mug. ☕️ Standard robots fail to distinguish "a cup" from "MY cup." Introducing VAP: A training-free framework that personalizes frozen VLAs using a few reference images. 🧵 [1/6] project: vap-project.github.io arXiv: arxiv.org/abs/2512.20014

Robots trained by behavior cloning learn which motion to reproduce, but not what the behavior was trying to achieve. Our new preprint "Act with Intent" fixes this by distilling behavior-level intent into the VLA action decoder. 🤖 📄 Paper: arxiv.org/abs/2608.23478 🌐 Project

arxiv.org

Act with Intent: Distilling Behavior Intent for...

Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was...

Excited to present our poster at #ICML2026 on Wednesday, July 8! 🤖 Bring My Cup! Personalizing Vision-Language-Action Models with Visual Attentive Prompting Poster Session, Hall A #816, 10:30 AM–12:15 PM We introduce the first personalization task for VLAs: manipulating your

1