๐ฅ#Legged_Robot_SLAM, State Estimation & Locomotion๐ฅ KAIST Urban Robotics Lab Ph.D
Joined December 2016
- Tweets14K
- Following1K
- Followers1.5K
- Likes27K
Pinned Tweet
During my graduate course, I had the opportunity to integrate a Mesh network-based Distributed Multi-Robot communication and evaluation system. This system utilized a synchronized time handler across robots to obtain reliable evaluation metrics. This system was also utilized in our latest work, Commerge.
youtu.be/LV_8_jZbHV8?si=PzITโฆ
![[IJRR 2026] Commerge: Communication-Efficient, Robust, and Fast LiDAR Map Merging Framework](https://pbs.twimg.com/card_img/2116449776854020096/9h2XORmV?format=jpg&name=800x419)
I visited one of the lava tubes in local area, and mapped it with iPhone 17 Pro.
Extremely challenging environment with visual degradation it was, but still succeeded on running odometry estimation with noisy and sparse LVI measurements
Jason Kim retweeted
์์ฆ์ ํ์ ์ด ๋ค๋ค ๋ง์ด ๋๊ธด ํ๋๋ผ๊ตฌ์ ์ ๋ง,,, ์... spin-spin.com/q/6ae34b972ce2โฆ
Jason Kim retweeted
Happy to share our SSIO! Very exciting direction to train IMU without labels!
๐ค ๐๐๐๐: ๐๐๐ฅ๐-๐๐ฎ๐ฉ๐๐ซ๐ฏ๐ข๐ฌ๐๐ ๐๐ง๐๐ซ๐ญ๐ข๐๐ฅ ๐๐๐จ๐ฆ๐๐ญ๐ซ๐ฒ
Since the first learned inertial odometry 10 years ago, every method has needed ground-truth labels.
It doesn't have to: SSIO learns from raw IMU, fully self-supervised ๐งต๐
ssio-imu.github.io
Jason Kim retweeted
Vision foundation models like DINO and SigLip now power much of robot learning.
But they are built for RGB. What about depth?
Introducing DeFM: a foundation model for depth images, built for robotics.
๐ de-fm.github.io
Weights, code, and demo are out today ๐งต
Jason Kim retweeted
๐คชmaintainer 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 ๐: @breadli428, Amin Abyaneh, @aahoussaini, Kirsty Ellis, @GlenBerseth, Marco Hutter, and Hsiu-Chin Lin from McGill, Mila, ETH Zรผrich & UdeM.
@mcgillu @Mila_Quebec @ETH_AI_Center @leggedrobotics @UMontreal
๐ arXiv:
arxiv.org/abs/2604.08780
๐ Project page (w/ videos):
modanesh.github.io/papers/qwโฆ
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/ACPโฆ
Promo Video ๐
Jason Kim retweeted
How can we represent whole-body interaction for humanoid robotsโcompactly enough for low-frequency planning, yet richly enough to preserve high-frequency feedback control?
Introducing FIBER! ๐ค
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!
๐ Website & Live Demo: fiberaa.github.io
๐ Paper: arxiv.org/abs/2610.24675
A thread on the ideas behind FIBER ๐งต๐
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 ๐งต๐
Jason Kim retweeted
ํ์ฌ 10๋
๋ค๋๋ฉด์ ๋ฐฐ์ด ๊ฒ ์๋ค. ๋๊ฐ๋ ์ฌ๋ ๋ง๋ก ๋ฐ๋ ํ์ฌ์์ผ๋ฉด ๋ค๋๋ ๋์ ์ด๋ฏธ ๋ฐ๋์๋ค.
๊ทธ๋์ ๋๋ ํด์ฌ ๋ฉด๋ด์์ ์ด ๋ง๋ง ํ๋ค. "ํ์ฌ๋ ๋ฐ๊ฟ ๊ฒ ์์ต๋๋ค. ์ ๊ฐ ์ ๋ง์์ ๋ฟ์
๋๋ค."
๋์ ํ ๊ฒ ์๋๋ค. ๋๋ฅผ ์งํค๋ ๊ฑฐ๋ค.
Jason Kim retweeted
๐ขOneCanvas (NeurIPS'26 Spotlight) Code Release๐ข
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!
๐ github.com/baranowskibrt/oneโฆ
๐ baranowskibrt.github.io/onecโฆ
![GitHub - baranowskibrt/onecanvas: [NeurIPS 2026 Spotlight] Official code and checkpoints for...](https://pbs.twimg.com/card_img/2116176042863779840/6dSrjkaU?format=jpg&name=800x419)
๐ข 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.
๐ baranowskibrt.github.io/onecโฆ
โถ๏ธ youtu.be/NIaHLB9gA7s
Great work by @baranowskibrt & @davech2y
Jason Kim retweeted
Weโll have two presentations at #IROS2026, featuring EFGCL and its extension, WARL!
Weโd love to see you there!
Details about the presentations are in the thread below โ
