RigidFormer: Learning Rigid Dynamics using Transformers

1. 05:00
   arXiv Robotics — research abstractsSignal

   RigidFormer: Learning Rigid Dynamics using Transformers

   This paper introduces RigidFormer, an object-centric Transformer model that learns high-fidelity rigid-body dynamics from mesh-free representations such as point clouds. The architecture uses compact anchors enriched with local vertex features via Anchor-Vertex Pooling to retain contact-relevant geometry without dense vertex-level interaction. It employs Anchor-based RoPE for spatial awareness while respecting the unordered nature of objects and enforces rigidity through differentiable Kabsch alignment. On standard benchmarks, RigidFormer outperforms or matches mesh-based baselines using point inputs, runs faster, generalizes to unseen point resolutions and across datasets, and scales to 200+ objects. The authors also demonstrate a preliminary extension to command-conditioned articulated bodies by treating body parts as interacting object-level components.

Current attention 3·Peak within the comparable range 4（2026-11-10 15:00 UTC）·Change within the comparable range over 24 hours –

The trend compares participants observed continuously; its scope may be narrower than current attention totals. Hover or click to explore hourly observations, or use the left and right arrow keys.