Donghu Kim

Manipulation RL Researcher @ Holiday Robotics

[Dongyoon Hwang @dongyoon_ai]

Introducing "3D HAMSTER", accepted to IROS 2026! 🎉

Hierarchical VLA planners draw waypoints in 2D, but robots act in 3D. Give the VLM a depth encoder, and it predicts metric 3D trajectories robots can execute.
Project page: https://t.co/5TiGDOmHLB
Paper: https://t.co/Rq3rIkUAP4

[Lukas Ziegler @lukas_m_ziegler]

South Korea's Holiday Robotics raised $105M Series A, largest single funding round for a Korean humanoid company

Their robot Friday has 64 DOF, with 40 dedicated to its hands. Tactile sensing, force-aware control, fully backdrivable joints, wheel-based mobility, hot-swappable batteries

Building full deployment stack in-house: Vision-Language-Skill (VLS) where vision and language define goals, reusable skills execute the work

[Donghu Kim @kty2396]

Huge thanks to the co-authors @hojoon_ai, @leeyngdo, @mpark1999, @kinam_0252, Takuma Seno, I Made Aswin Nahrendra, @seheemax, @DPalenicek, Florian Vogt, @DanicaKragic, @Jan_R_Peters, Jaegul Choo, and @holidayrobots who made room for this work, and @younggyoseo for the discussion!

[Donghu Kim @kty2396]

FlashSAC won the Outstanding Paper Award at RSS 2026 🎉

We got off-policy RL fast and stable enough to beat PPO and FastTD3 across 60+ tasks and 10 simulators with minimal tuning!

TL;DR: If you're working on dexterous manipulation, just try FlashSAC!

https://t.co/hYlhuLRTXn

[Donghu Kim @kty2396]

Still a little stunned how everything turned out. We ourselves were unsure whether off-policy (or even RL) is the right choice for complex robotic tasks; seems like we proved ourselves wrong. Excited to see what RL can further achieve in dexterous manipulation!

[Robotics: Science and Systems @RoboticsSciSys]

#RSS2026 Awards
🧵1. Outstanding paper
FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control
Donghu Kim, Youngdo Lee, Minho Park, Kinam Kim, Takuma Seno, I. Made Aswin Nahrendra, Sehee Min, Daniel Palenicek, Florian Vogt, Danica Kragic, Jan Peters, Jaegul Choo, Hojoon Lee
https://t.co/fpdIFY8io1

[Byungkun.Lee @Byungkun_Lee]

Introducing "See like a Robot"🤖
Robot data spans diverse camera viewpoints, making learning harder. Give a VLA robot-centric pointmaps, and it performs better with one extra encoder + one element-wise addition.
Project page: https://t.co/k5uBJOUBYa
Paper: https://t.co/ff2pwLLxxl

[Hojoon Lee @hojoon_ai]

We scaled off-policy RL to sim-to-real.
To our knowledge, FlashSAC is the fastest and most performant RL algorithm across IsaacLab, MuJoCo Playground, and many more, all with a single set of hyperparameters.
Project page: https://t.co/uaTcOoYtjt
Paper: https://t.co/PLu6ZGRKuB