Pinned Tweet
What if physics simulation worked like video generation?
We introduce PhysLDM: a single latent diffusion model that predicts entire trajectories of diverse high-resolution 3D deformable objects at once and generalizes zero-shot. A step toward physical world models. 🧵
Yu Zhang @ NeurIPS 2026 retweeted
Announcement:
I have indeed been diagnosed with the disease that everyone fears, but it was caught at a very early stage.
I have already scheduled my surgery, and I thank everyone for your concern.
Honestly, I no longer care about so-called money or prestige. I just want to do what I can within my academic field and work on AI things that bring me joy (the doctor says this might also help with my recovery).
I have let go of all past interpersonal entanglements, as well as others' prejudices and discriminations against me.
In the face of life and death, these things are so insignificant. It is this illness that has made me rethink the meaning of life and death: one should not live for the approval of others, but persist for the things we love and the people who love us.
I will continue to share because it is a meaningful thing to do.
4️⃣ Differentiable, unlocking hard inverse problems
With end-to-end and higher-order gradients, PhysLDM efficiently and accurately recovers Young's modulus from a single trajectory, and solves second-order design tasks that are prohibitive for existing differentiable simulators.
1
2
290
🌐 Project page: yucrazing.github.io/physldm
📄 arXiv: arxiv.org/abs/2610.07609
RTs & feedback welcome 🙌
Our ClothTransformer dataset has released now!
🤗 huggingface.co/datasets/YuCr…
493K frames penetration-free cloth simulation (2,056 trajectories, ~33GB) across 3 scenarios:
🪂 Free-fall collision onto diverse objects
👕 Garments on animated human bodies
🤖 Robotic cloth manipulation
Sharing our recent work — ClothTransformer 🧵
We propose a unified Transformer-based neural cloth simulator that solves the "one model per scenario" problem. One single model handles diverse cloth simulation scenarios — useful for digital humans, embodied AI, games & VFX.
3️⃣ Penetration-free data + differentiable CCD
We built a ~493.4k-frame high-quality multi-scenario penetration-free cloth dataset, which supports a differentiable Continuous Collision Detection (CCD) module to effectively suppress penetration artifacts.
1
1
453
🌐 Project page: yucrazing.github.io/clothtra…
📄 arXiv: arxiv.org/abs/2605.27852
RTs & feedback welcome 🙌
2
164


