Research map / Vision-language and multimodal / Robot action and navigation
PointWorld: Scaling 3D World Models for In-The-Wild Robotic Manipulation
Wenlong Huang, Yu-Wei Chao, Arsalan Mousavian, Ming-Yu Liu, Dieter Fox, Kaichun Mo, Li Fei-Fei
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace GBND backbone with PTv3-1B | PointWorld · DROID test set | l2 mover: 0.0390 -> 0.0312 | Improved |
| replace GBND backbone with PTv3-1B | PointWorld · DROID test set | l2 static: 0.0066 -> 0.0056 | Improved |
| scale PTv3 from 50M to 1B parameters | PointWorld · DROID test set | l2 mover: 0.0331 -> 0.0312 | Improved |
| use Transformer instead of PointNet++ | PointWorld · DROID test set | l2 mover: 0.0368 -> 0.0339 | Improved |
| use SparseConv instead of GBND | PointWorld · DROID test set | l2 mover: 0.0390 -> 0.0396 | Hurt |
| use PointNet instead of GBND | PointWorld · DROID test set | l2 static: 0.0066 -> 0.0084 | Hurt |
| add movement weighting, uncertainty regularization and Huber loss (ablation) | PointWorld · DROID test set | l2 mover | Improved |
| add frozen DINOv3 features to scene points (ablation) | PointWorld · DROID test set | l2 mover | Improved |
| use gripper-only point flows for actions instead of whole-body flows or low-dim actions (ablation) | PointWorld · DROID real-world | l2 mover | Improved |
| use chunked prediction in training and inference instead of autoregressive rollout (ablation) | PointWorld · DROID test set | l2 mover | Improved |
| train with random camera counts instead of fixed counts (ablation) | PointWorld · DROID test set | l2 mover | Improved |
| add more training data and model capacity (5% to 100% data, 50M to 1B) | PointWorld · DROID test set | l2 mover | Improved |
Closest papers by meaning
- PointZero: 3D Point Track Completion for Learning Transferable 3D Dynamics (NeurIPS 2026 · similarity 0.90)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.89)
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.88)
- GravMAD: Grounded Spatial Value Maps Guided Action Diffusion for Generalized 3D Manipulation (ICLR 2025 · similarity 0.88)
- World Motion Models: Flexible Sequence Modeling of SE(3) Trajectories (NeurIPS 2026 · similarity 0.88)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
