Hojune Kim

Robotics @Stanford
prev @DLR_en

[Hojune Kim @HojuneKim]

What if a phone scan is all you need to teach a humanoid a new skill, one that generalizes to scenes it's never seen?

Introducing 🦿LEGS🦿: a photorealistic loco-manipulation simulator. No teleop; policies deploy zero-shot on a Unitree G1 🤖

https://t.co/Ubfep7hP7l 👇

[Hojune Kim @HojuneKim]

@AmyNoteApp That's exactly what the SAM-3D baseline controls for! Same contacts, same motion, only the rendering differs (mesh only, no 3DGS). Averaged across all experiments, it roughly halves success. So with contact geometry held fixed, the win was the rendering.

[Hojune Kim @HojuneKim]

What if a phone scan is all you need to teach a humanoid a new skill, one that generalizes to scenes it's never seen?

Introducing 🦿LEGS🦿: a photorealistic loco-manipulation simulator. No teleop; policies deploy zero-shot on a Unitree G1 🤖

https://t.co/Ubfep7hP7l 👇

[Hojune Kim @HojuneKim]

[11/11] Huge thanks to my collaborators at @StanfordMSL: Timothy Chen, @JiankaiSun, Lars Osterberg, @QianzhongChen, @ke_wang123, and @MacSchwager to make all this work! 🥳

Check out the links below:
📄 Paper: https://t.co/OMBX3xCu94
🌐 Website: https://t.co/Ubfep7hP7l

[Hojune Kim @HojuneKim]

[10/11] Under the hardest object + scene shift, two takeaways 📊 (a) the photorealism edge holds: LEGS-aug still beats mesh-only SAM3D-aug at matched 200 episodes. (b) augmentation beats scale: just 50 re-rendered episodes outperform 200 default-only.