Model family

Lift, See, Act (LSA)

Participating organizations

Version scope and roles for these associations await verification.

View evidence

Established by researchers at South China University of Technology, Lift, See, Act pretrains visual representations for robot manipulation from human video. It aligns intermediate features with two 3D teacher models, uses depth reconstruction and hand-region guidance, and learns to predict future 3D wrist motion. The authors evaluated the learned encoder in simulation and on a Franka Research 3 arm.

Overview

Model type
Component
Architecture
Transformer
Listed versions
Lift, See, Act (LSA)
Related robots
1

Papers & reports (1)

Lift, See, Act: Hierarchical Robot Policy Pretraining with 3D Foundation Models

Paper available

arXiv: 2611.20481
Authors & affiliations (5)

Author order: Original order

  1. Yiyuan Ge

    School of Electronic and Information Engineering, South China University of Technology

  2. Changxing DingCorresponding author

    School of Electronic and Information Engineering, South China University of Technology

  3. Ziyu Hao

    School of Electronic and Information Engineering, South China University of Technology

  4. Zijie Zheng

    School of Electronic and Information Engineering, South China University of Technology

  5. Xiangmin Xu

    School of Electronic and Information Engineering, South China University of Technology

First published
2026-11-16
Latest paper revision
v1 · 2026-11-16
Venue
NeurIPS 2026

Covered model versions

Paper ↗PDF ↗

Official resources

Submit a source or correction