For everyone interested in precise 馃摲camera control 馃摲 in transformers [e.g., video / world model etc]
Stop settling for Pl眉cker raymaps -- use camera-aware relative PE in your attention layers, like RoPE (for LLMs) but for cameras!
Paper & code: liruilong.cn/prope/View quoted post on X
I'm super excited to announce mjlab today!
mjlab = Isaac Lab's APIs + best-in-class MuJoCo physics + massively parallel GPU acceleration
Built directly on MuJoCo Warp with the abstractions you love.View quoted post on X
Humanoid motion tracking performance is greatly determined by retargeting quality!
Introducing 饾棦饾椇饾椈饾椂饾棩饾棽饾榿饾棶饾椏饾棿饾棽饾榿馃幆, generating high-quality interaction-preserving data from human motions for learning complex humanoid skills with 饾椇饾椂饾椈饾椂饾椇饾棶饾椆 RL:
- 5 rewards,
- 4 DR terms,
- Proprio. ONLY,
- NO history/curriculum.
Ready for agile, human-like 馃? (Best with 馃帶)
馃敆 omniretarget.github.io 馃帴
1/9View quoted post on X
