🔥#Legged_Robot_SLAM, State Estimation & Locomotion🔥 KAIST Urban Robotics Lab Ph.D
Joined December 2016
- Tweets13.9K
- Following1K
- Followers1.5K
- Likes27K
Jason Kim retweeted
🤪maintainer of GTSAM, Gatech phd, auto rejected for a state estimation role.
Hiring in robotics is so funny. I applied to @shieldaitech for a state estimation role, especially since the job description listed being familiar with libraries like GTSAM. You'd think I would be a great fit.
Got a rejection email in less than a week without any interviews. :)
Jason Kim retweeted
This was a fun collaboration with an amazing team 🙌: @breadli428, Amin Abyaneh, @aahoussaini, Kirsty Ellis, @GlenBerseth, Marco Hutter, and Hsiu-Chin Lin from McGill, Mila, ETH Zürich & UdeM.
@mcgillu @Mila_Quebec @ETH_AI_Center @leggedrobotics @UMontreal
📄 arXiv:
arxiv.org/abs/2604.08780
🌐 Project page (w/ videos):
modanesh.github.io/papers/qw…
Jason Kim retweeted
An early video of testing state estimation solutions for box climbing. Livox+dlio vs odin1
With @JayHe748646 @KehanWen170077 @leggedrobotics , we push humanoid mobility to next frontier.
Achieved by a generalist contact planner + a generalist tracker + a generalized neural mapping, warmstarted from few human data.
Website: leggedrobotics.github.io/ACP…
Promo Video 👇
Jason Kim retweeted
How can we represent whole-body interaction for humanoid robots—compactly enough for low-frequency planning, yet richly enough to preserve high-frequency feedback control?
Introducing FIBER! 🤖
FIBER learns a unified latent interface between high-level planning and whole-body control, enabling a humanoid to acquire diverse interaction behaviors from heterogeneous experts through a pretrained, frozen Behavior Foundation Model (BFM).
With a single planner and BFM, we demonstrate heavy-door pushing, box carrying, object grasping, controllable navigation, and more!
🌍 Website & Live Demo: fiberaa.github.io
📄 Paper: arxiv.org/abs/2610.24675
A thread on the ideas behind FIBER 🧵👇
Jason Kim retweeted
No prior data. No rewards. No pretrained models.
Can an RL agent learn to solve extremely long-horizon tasks (e.g., 5-cube stacking) from scratch?
We show it can! C-Slerp learns hyperspherical contrastive representations and infers waypoints from them in closed form 🧵👇
Jason Kim retweeted
📢OneCanvas (NeurIPS'26 Spotlight) Code Release📢
What if a 2D VLM could see the whole scene in one view?
We reproject features from every frame into a single occlusion-free canvas to make a stock VLM reason in 3D!
🔗 github.com/baranowskibrt/one…
🌐 baranowskibrt.github.io/onec…
📢 OneCanvas: 3D Scene Understanding via Panoramic Reprojection
We extract features from video frames and reproject them into one occlusion-free view of the whole scene that a 2D VLM reads just like a normal image. We can center this view on any viewpoint, including an agent's own pose for situated reasoning.
The same projection lets us create spatial training tasks with no human annotation, solvable only by reasoning over the 3D positions of real object features placed on an otherwise empty canvas.
The result is a stock 2D VLM that reasons in 3D, setting a new state of the art across spatial benchmarks at far less compute.
🌐 baranowskibrt.github.io/onec…
▶️ youtu.be/NIaHLB9gA7s
Great work by @baranowskibrt & @davech2y
Jason Kim retweeted
We’ll have two presentations at #IROS2026, featuring EFGCL and its extension, WARL!
We’d love to see you there!
Details about the presentations are in the thread below ↓
Jason Kim retweeted
⚡ One policy, millions of embodiments, over 200 robot models. Can we add yours?
We're building γ₀, a generalist RL policy for motion control trained across millions of randomized embodiments derived from a growing collection of more than 200 robot models.
Jason Kim retweeted
Our LT-Mem paper won the Best Paper Award at IROS 2026 — 1 out of 1,933 papers!
Huge congratulations to Yumin Lee and Hyoseok Ju. Proud of our APRL team at DGIST!
lt-mem.github.io/
#IROS2026 #Robotics #SpatialIntelligence #DGIST #APRL
Jason Kim retweeted
토큰 맥서들의 결말
– 로컬 LLM이 답이라며 고가의 하드웨어 사재기
– 에이전트를 위한 에이전트를 만들다가 열화된 결과물에 주화입마
– 폭증하는 코드를 감당 못해 개인 비서 만들다 목적을 상실
– 결과물이 안나오니 그저 과정을 소셜에 전시하고 파는 일에 집중
Jason Kim retweeted
Had a small singularity moment at the lab today 😃
Today, my agents turned the robot on, ran a mission for a few minutes, stopped it, pulled the logs, analyzed them, changed the code, deployed it, and ran it again. Then they kept going. Test after test, deployment after deployment.
The entire engineering loop is now closed 🙂 This is where all the work we’ve put into Ayg as a platform comes together. We’ve built the hardware and software as one connected system, and the new autonomy layer builds on that foundation, making it much easier for AI agents to work with the robot.
I set the objective. The agents handled the iteration loop with 0 human intervention.
This is one of the doors layered autonomy opens: agents can now test their changes on the physical robot and keep iterating without me sitting between the software and the hardware.
I knew this was possible. I’ve been experimenting with it for a long time, but watching it happen still feels strange. I’m just sitting there while the robot turns on, walks for five minutes, stops, and a few minutes later starts again with updated code.
It’s surreal~
This is something we call layered autonomy. Ayg can now run autonomously for hours.
I had a PoC of this architecture 6 months ago, and it worked, but nowhere near as well as it does today. The missing piece of the puzzle was Jev.
Jev is the tactical layer in our autonomy stack. In the original PoC, this layer and the modules around it were specialized RL policies.
And to be honest, I still think the long-term solution is to have local, specialized RL policies for the tasks we use Jev for today. But Jev gave us a way to make the architecture work now, and made me much more confident that the original direction was right.
Huge kudos to the @typesafeai team.
The autonomy graph finally feels complete, and the possibilities it opens up are pretty amazing
Jason Kim retweeted
単一のファクタグラフ上でLiDAR-IMUオドメトリとオンラインIMU運動モデル学習を統合する手法を提案しています。LiDARが強く効く場面ではIMU運動モデルの学習を行いつつ、LiDARが縮退する環境では学習したIMU運動モデルを利用して頑健な自己位置推定を行うことができます。
[#1919] Kazushi Fujimoto (@6LoMvsNUZcH4c2Y), Tightly Coupled LiDAR-IMU Odometry with Online Adaptation of a Learning-Based IMU Motion Model
Paper: staff.aist.go.jp/k.koide/ass…
Video: youtu.be/udsEVMJz7Ag
Link: 2026.ieee-iros.org/program/c…
Jason Kim retweeted
多様なロボットの身体構造をグラフ表現で記述し、統一的に運動予測を行うモデルを提案しています。車輪や脚のようなまったく違う構造のロボットの運動も単一のネットワークで扱うことができ、実ロボットにもfew-shotでの適応が可能です。
[#1824] Taku Okawara (@NNUNNTT), KING: Embodiment-Aware Kinematic Graph Neural Network for Unified Motion Representation of Legged and Wheeled Robots
Paper: arxiv.org/pdf/2608.01015
Video: youtube.com/watch?v=UVXCN13b…
Page: smrg-aist.github.io/king_pro…
Link: 2026.ieee-iros.org/program/c…
Jason Kim retweeted
生成モデルを用いてスパースなLiDAR反射強度画像のdensificationを行い、それとカメラ画像のマッチングを行うパイプラインを提案しています。明示的なLiDAR・カメラペアデータを用いることなく、頑健かつ高速なセンサフュージョンを実現しています。
[#1949] Reon Tabata (@reonrl), Image-To-Point Cloud Registration Made Easy with Rectified Flow-Based LiDAR Upsampling
Paper: arxiv.org/pdf/2607.14639
Video: youtube.com/watch?v=r8I94aCC…
Page: smrg-students.github.io/iros…
Link: 2026.ieee-iros.org/program/c…
