58 motions to start: dances, waves, bows and styled walks, all retargeted to K1. We'll keep maintaining and extending it.
K1 Moves: huggingface.co/datasets/Kyry…
SONIC-K1: kyry.me/sonic-k1/
Kyr retweeted
Isaac Lab 3.0 lets you choose the physics and rendering backends that best fit your robot learning workload, as well as different visualization approaches to see what your policy looks like while training.
For example, the SONIC whole-body controller from NVIDIA runs across different physics engines and transfers to real hardware, demonstrating policy robustness across simulation backends and the real world.
(3/7)
GitHub: github.com/Kyk0/sonic-k1
Replying to @kyry_D
Weights and code: kyry.me/sonic-k1
Next: the real K1, whole-body VR teleop, and video to robot.
*On the bigger Unitree G1, as reported in the SONIC paper.
Weights and code: kyry.me/sonic-k1
Next: the real K1, whole-body VR teleop, and video to robot.
*On the bigger Unitree G1, as reported in the SONIC paper.
Weights and code: kyry.me/sonic-k1
Next: the real K1, whole-body VR teleop, and video to robot.
*On the bigger Unitree G1, as reported in the SONIC paper.
Humanoids are starting to surpass human physical capabilities in some areas. I think we'll soon have a new regulations on how powerful they can be
Kyr retweeted
I'm shocked by the lack of ambition of recent robot companies. The models keep on getting on better, but the robots stay the same.
The honest truth is that you can build a humanoid in ~2 months.
Code and guide: github.com/Kyk0/isaaclab-h10…
MIT / CC BY 4.0. Tested with Isaac Lab 2.x on H100 NVL. Community-tested, not NVIDIA-supported.
If you try it on A100, H200 or B200, send a GPU report and I'll add it with credit.
You can watch the policy in your browser.
Record the robot's link poses during a headless rollout, then replay them with viser. Your browser handles the drawing, so the server doesn't need to render.
Try it: kyry.me/isaaclab-h100/#brows…










