People grow.
Why not robots?

Robots that handle objects they have never seen, with bodies they have never used, and get better the more they are used.

GROW Lab builds General-purpose Robots for the Open World.
We study how robots perceive, act and keep learning in the real world, from grippers and dexterous hands to humanoids.

Join the lab Research questions
Grippernew objectsArmnew scenesDexterous handnew bodiesHumanoidkeeps learning
Research questions

Why aren't robots in our homes yet?

Because the real world keeps changing: new objects, new robot bodies, and everything after deployment. We ask one question for each.

Open-World Manipulation

Can a robot handle an object it has never seen, with no CAD model?

  • Model-free 6D pose and scale
  • Novel and fast-moving objects
  • Skills from one demonstration

Related: Any6D, Event6D, GraspClutter6D

Cross-Embodiment

Can the same skill run on a new gripper, hand or camera without retraining?

  • One grasp model, many grippers
  • Markerless hand-eye calibration
  • Viewpoint-robust VLA policies

Related: XGrasp, GraspGraphNet

Lifelong Robot Learning

Can a deployed robot keep improving from its own successes and failures?

  • Test-time and domain adaptation
  • Automatic 6D labeling of collected data
  • Failure detection and recovery

Related: TTA-COPE, UDA-COPE

  1. Open-WorldSee anything
  2. Cross-EmbodimentAct with any body
  3. LifelongLearn from every try

↻Repeat, and the robot grows.

Where it lands

Research Highlights

Robots meet the real world

Novel objects, clutter, fast motion, new grippers and hands, on real robots.

Any6D · CVPR 2025Grasping novel objects with Any6DOne RGB-D anchor frame, no CAD model, pose and size fed straight to the grasp planner.
Event6D · CVPR 2026Fast-moving object tracked with an event cameraPose keeps up where frame cameras blur.
TTA-COPE · CVPR 20236D pose that adapts to a new sceneCategory-level pose keeps improving at test time, with no labels from the new scene.
GraspClutter6D · RA-L 2025Grasping in clutterPicks diverse objects one by one from a cluttered bin into an empty box.
OngoingVLA for dynamic objectsOur ongoing vision-language-action project for dynamic manipulation.
DeLTaMake green liquid in the cylinderPours yellow and blue from novel transparent bottles, guided by one demonstration.
XGraspOne grasp model, many grippersSwap the gripper, keep the model. Same weights, no retraining.
GraspGraphNetDexterous grasps on a real handGraph-structured grasps for multi-fingered hands, on objects of any shape.

Publications

* equal contribution, † corresponding. Full list on taeyeop.com and Google Scholar.

2026

2025

2024

2023

2022

2021

People

Taeyeop Lee

Taeyeop Lee

Principal Investigator · Senior Researcher, AI Robot Institute, KIST · Ph.D., KAIST

Taeyeop is a Senior Researcher at the AI Robot Institute, KIST. He received his Ph.D. and M.S. from KAIST and his B.S. from Kwangwoon University. He has been fortunate to collaborate with NVIDIA Research, Meta and Adobe Research, and is a recipient of the Presidential Award of Korea.

Homepage Google Scholar GitHub LinkedIn

Founding members wanted

M.S. and Ph.D. studentsBuild our first robots with us.
Interns and undergraduatesReal robots from day one.
Post-docsLead your own direction.

News

Join us

Do world-class robot AI with world-class collaborators.

We work with researchers at NVIDIA, Google, Meta and KAIST, and aim every project at the top venues in robotics and AI. Join us to bring VLA models, robot learning, dexterous hands and humanoids into the real world.

Apply formthen send an email to taeyeop.trevor@gmail.com to let us know you applied.
KIST official calls

Student recruiting notices (M.S./Ph.D., Jan. & Aug.) · Employment announcements (interns, post-docs)

Collaborations

Institutions we have worked and published with.

IndustryNVIDIAGoogleMetaAdobeNAVER LABSKRAFTON
AcademiaCarnegie Mellon UniversityKAISTTUMGISTKIMM