Pinned post
Introducing PointZero—a 3D world model pre-trained without robots.
Dexterous manipulation requires understanding diverse 3D dynamics, but current approaches rely on expensive robot data. So how can we pre-train a 3D dynamics model without it?
PointZero introduces a simple idea: learning to complete 3D point tracks yields transferable 3D dynamics!
Our model achieves SOTA results on:
🔥 Zero-shot 3D dynamics
🔥 Action-conditioned 3D dynamics (post-training)
🔥 Imitation learning (post-training)
Details and links 👇 1/6
Dexterous manipulation requires understanding diverse 3D dynamics, but current approaches rely on expensive robot data. So how can we pre-train a 3D dynamics model without it?
PointZero introduces a simple idea: learning to complete 3D point tracks yields transferable 3D dynamics!
Our model achieves SOTA results on:
🔥 Zero-shot 3D dynamics
🔥 Action-conditioned 3D dynamics (post-training)
🔥 Imitation learning (post-training)
Details and links 👇 1/6
0:43
