DREAM Lab.
EN KO

What we do

Research

From arm manipulation to mobile manipulators and humanoids — robots that work autonomously in dynamic environments.

01

Motion Planning

Planning feasible grasps and motions under kinematic, closed-chain, and task constraints — from single arms to whole-body systems.

  • Grasp Planning
  • Whole-body Motion Planning
  • Constrained Motion Planning
  • GPU-Parallel Planning

Related publications

  • PATACON: GPU-Parallel Tangent-Bundle RRT-Connect for Constrained Motion Planning Preprint 2026arXiv
  • Connectivity-Aware Representations for Constrained Motion Planning via Multi-Scale Contrastive Learning IEEE ICRA 2026arXiv
  • A Constrained Motion Planning Method Exploiting Learned Latent Space for High-Dimensional State and Constraint Spaces IEEE/ASME T-MECH 2024Paper
  • Motion Planning for Closed-Chain Constraints Based on Probabilistic Roadmap With Improved Connectivity IEEE/ASME T-MECH 2022Paper

02

Physical AI

Learning manipulation skills from data — imitation learning, vision-language-action models, teleoperation for data collection, and post-optimization of learned policies.

  • Post-optimization Algorithm
  • Vision-Language-Action Models
  • Imitation Learning
  • Teleoperation
  • Dataset Augmentation

Related publications

  • HITMEN: A Hybrid Interface for Humanoid Teleoperation with Multimodal Sensing and Exoskeleton Control Manuscript 2026Project
  • SNU-Avatar Haptic Arm: A Hybrid QDD/DD Operator Interface for Teleoperation IEEE Access 2026Paper
  • LiPo: A Lightweight Post-optimization Framework for Smoothing Action Chunks Generated by Learned Policies IJCAS 2025Paper
  • Intuitive and Interactive Robotic Avatar System for Tele-Existence: TEAM SNU in the ANA Avatar XPRIZE Finals IJSR 2024Paper

03

Humanoids

Whole-body manipulation and bipedal walking for humanoid robots working in real environments.

  • Whole-body Manipulation
  • Bipedal Walking

Related publications

  • HITMEN: A Hybrid Interface for Humanoid Teleoperation with Multimodal Sensing and Exoskeleton Control Manuscript 2026Project
  • Mechanical Design of a Giant Human Riding Biped Humanoid Robot, Method II IJHR 2026Paper
  • Dual-Channel EtherCAT Control System for 33-DOF Humanoid Robot TOCABI IEEE Access 2023Paper
  • Weighted Hierarchical Quadratic Programming: Assigning Individual Joint Weights for Each Task Priority ISR 2022Paper

Projects

Funded Projects

Swarm Humanoid
K-Moonshot Project ↗ OngoingNRF

Swarm Humanoid

Development of Autonomous Task Distribution and Physical Collaboration Framework for Multi Heterogeneous Humanoid Robots

FundingNational Research Foundation of Korea (NRF) · K-Moonshot ProjectPeriod2026.04.01 ~ 2030.12.31

A perception–understanding–collaboration platform in which multiple humanoids interpret natural-language (LLM) commands, share sensor data for swarm perception, and use agentic AI to decompose, distribute, and execute tasks in logistics, manufacturing, and disaster settings — validated in the field.

Junho An, Gahyun Oh, Jinwoo Lee, Dongwoo Son, Chanwoo Seon

  • Swarm Robotics
  • Agentic AI
  • LLM
  • Humanoid
OngoingNRF

Constraint-aware VLA

Development of a Constraint-Aware Vision-Language-Action Model for High-Precision Robotic Manipulation

FundingNRF · Ministry of Science and ICTPeriod2026.09.01 ~ 2029.08.31

A constraint-reasoning Vision-Language-Action model (CA-VLA) that understands language instructions and visual input while explicitly reasoning about the constraints a task imposes. Manipulation is decomposed into approach, alignment, contact, and insertion phases whose geometric, contact, force/torque, ordering, and safety constraints are mapped to contact points, alignment axes, tolerances, and keep-out regions in the robot frame; model-based control and planning then correct the VLA's actions to improve success and stability in high-precision tasks such as peg-in-hole, rotational assembly, and bimanual manipulation.

Junho An, Gahyun Oh, Dongwoo Son

  • VLA
  • Physical AI
  • Constraint Reasoning
  • High-Precision Manipulation
OngoingROBOTISMOTIR

BGF Logistics Automation

Development of AI Foundation Model-Based Humanoid Robots Specialized for Distribution Processes

FundingKEIT · Ministry of Trade, Industry and EnergyPeriod2025.09.01 ~ 2028.12.31

Developing a humanoid system (AI Worker) specialized for distribution processes on top of AI foundation models, and applying VLA and Physical AI to automate manual work in real distribution centers.

Chanwoo Seon, Jinwoo Lee, Gahyun Oh, Dongwoo Son, Hwang Young Sang

  • VLA
  • Physical AI
  • AI Worker
  • Logistics Automation
OngoingROBROS

Humanoid Package Handling

Development of Humanoid Physical AI for Autonomous Handling of Diverse Packages

FundingSeoul Metropolitan Government (AI Technology Commercialization Program)Period2026.07.01 ~ 2027.06.30

Physical AI that lets a humanoid move mixed, irregular packages (boxes, fresh totes, and soft bags up to 6 kg) from a workbench to a conveyor for over 60 minutes without human intervention. The project integrates an upgraded IGRIS C platform, RL-based whole-body balance control, constraint-based bimanual motion planning, and learned grasping that generalizes to unseen packages, validated on a logistics testbed and in real-world PoCs.

Dongwoo Son, Chanwoo Seon, Junho An, Gahyun Oh, Jinwoo Lee

  • IGRIS C
  • Physical AI
  • Bimanual Manipulation
  • Whole-body Control
  • Logistics Automation
Auto-bagger Automation
CompletedROBROS

Auto-bagger Automation

Early Demonstration and Commercialization Study of AI Bipedal Humanoids

FundingKIRIA · Ministry of Trade, Industry and EnergyPeriod2025.09.01 ~ 2025.12.31

AI bipedal humanoids were deployed to three universities, each running field demonstrations built around a partner company's work environment to validate early performance and safety, assess commercialization potential, and derive implications for safety regulation.

Dongwoo Son, Junho An, Chanwoo Seon, Jinwoo Lee

  • IGRIS C
  • Field Demonstration
  • Sim-to-Real
  • Commercialization