Robotics · 2026-10-19
ORD-WM: Learning Object-Centric Rigid Dynamics for 3D World Modeling in Robotic Manipulation
Yueyang Weng, Xiaopeng Zhang, Yuxi Ou, Yongjin Mu, Chengzhi Zhao, Yanjie Li
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
ORD-WM is a 3D world model for robotic manipulation that learns object-centric rigid dynamics. It uses an SDF-supervised vector-neuron VAE to encode each object point cloud as geometry-aware 3D vector latents, then predicts object-wise SE(3) motion increments conditioned on robot action. Instead of unconstrained latent prediction, it analytically applies rigid transformations to latent states, yielding stable geometry-preserving multi-step rollouts and efficient closed-loop MPC.
Why it matters
The authors target drift and inconsistent rigid-object geometry in multi-step rollouts. By enforcing rigid object motion through analytically applied SE(3) updates in a geometry-aware latent space, ORD-WM aims to preserve object shape/identity while still updating pose, enabling more reliable model-based control from predicted object states.
Method
- SDF-supervised SDF-VNVAE encodes each object point cloud into SO(3)-equivariant 3D vector-feature latents and supports rotation-aware SDF decoding.
- Action-conditioned transformer dynamics predicts object-wise SE(3) translation increments and 6D rotation representations for floating objects.
- Latent rollout applies predicted object-wise SE(3) rigid transformations analytically to the current object latents (geometry preserved), while fixed objects remain unchanged.
Limitation
The paper does not state a specific limitation in the provided text.
Abstract (from arXiv)
World models provide robots with a predictive mechanism for reasoning about how actions change object states in manipulation tasks. However, existing image-based and dense 3D world models do not explicitly impose object-level rigid motion, leaving rigid-object geometry preservation to be learned implicitly and making multi-step rollouts susceptible to shape drift or object-level motion inconsistency. To address this, we propose ORD-WM, a 3D world model that learns object-centric rigid dynamics for robotic manipulation. ORD-WM learns a vector-neuron variational autoencoder supervised by signed distance fields, which encodes each object point cloud as a geometry-aware latent representation composed of 3D vector features, and uses an action-conditioned dynamics model to predict object-wise SE(3) motion increments. Rather than generating unconstrained next latent states, ORD-WM analytically applies predicted object-wise rigid transformations to the current latent states, separating persistent object geometry from pose changes. Experiments show that ORD-WM enables stable, geometry-preserving multi-step prediction and supports efficient real-time replanning for closed-loop model predictive control in simulated and real-world manipulation scenarios.
Related papers
- PointWAM: 3D World Action Modeling for Dexterous Robotic Manipulation
- Relational Geometry Forcing for World Action Models
- UNITAS: A 3D-Native World Action Model for Embodied Manipulation
- SkeleWAM: Skeleton World-Action Modeling for Efficient Robotic Manipulation
- TrajWAM: An Efficient Trajectory-Guided World Action Model for Dynamic Robotic Manipulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-16 · Indexed by Pub-AI: 2026-10-19 UTC
ORD-WM: Learning Object-Centric Rigid Dynamics for 3D World Modeling in Robotic Manipulation
Preprint posted: 2026-10-16 · Original record ↗
- arxiv
- 2610.23156
Explicit source versions (1)
- v1 · 2026-10-16
- Preprint available
Source date: 2026-10-16 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World models provide robots with a predictive mechanism for reasoning about how actions change object states in manipulation tasks. However, existing image-based and dense 3D world models do not explicitly impose object-level rigid motion, leaving rigid-object geometry preservation to be learned implicitly and making multi-step rollouts susceptible to shape drift or object-level motion inconsistency. To address this, we propose ORD-WM, a 3D world model that learns object-centric rigid dynamics for robotic manipulation. ORD-WM learns a vector-neuron variational autoencoder supervised by signed distance fields, which encodes each object point cloud as a geometry-aware latent representation composed of 3D vector features, and uses an action-conditioned dynamics model to predict object-wise SE(3) motion increments. Rather than generating unconstrained next latent states, ORD-WM analytically applies predicted object-wise rigid transformations to the current latent states, separating persistent object geometry from pose changes. Experiments show that ORD-WM enables stable, geometry-preserving multi-step prediction and supports efficient real-time replanning for closed-loop model predictive control in simulated and real-world manipulation scenarios.