PhysLDM reconstructs deformable trajectories at 2.48mm precision with 78x token compression
jiqizhixin · x · 2026-11-21
Researchers from Nanyang Technological University and Shanghai AI Laboratory introduce PhysLDM in an arXiv preprint on volumetric deformable simulation.
- Spatiotemporal autoencoder: separates rigid motion from local deformation and compresses entire trajectories, reconstructing deformable-body trajectories at 2.48mm precision on meter-scale scenes with up to 78-fold token compression.
- Diffusion prediction: a diffusion model (Flow-DiT, per the paper diagram) predicts the trajectory in latent space from the initial state, making joint prediction tractable for meshes whose full space-time representation would be too costly.
- Generalization: trained on simulated Objaverse objects, transfers to unseen GSO and Toys4K objects; the paper also contrasts diffusion with deterministic regression, which can average away complex dynamics.
More from Embodied
- Open-source humanoid Asimov starts wheeled variant on day 389 of development — freelerobot · 2026-11-21
- USC's Heave-Ho trains Unitree G1 humanoids to collaborate, scales 2 to 10 robots — CyberRobooo · 2026-11-21
- Wanting a full self-driving Tesla RV: futuristic Cybertruck-style camper concept — PaulYacoubian · 2026-11-21
- Open-source autonomous driving book's Chapter 4 compares Pure Pursuit, Stanley, rear-wheel feedback — 4310sy · 2026-11-21
- SAFI teaches VLA robots to stop safely mid-task, cutting takeover success to 1.3% — DM2-ND · 2026-11-21
- Berkeley open-sources Specialist: 3D-printed gripper for robot data, runs on real arms — SongShuran · 2026-11-21