arXiv:2603.02263v3 Announce Type: replace-cross Abstract: World models compress rich sensory streams into compact latent codes that anticipate future observations. We let separate agents acquire such...
arXiv:2605. 28865v2 Announce Type: replace-cross Abstract: What does a world model learn from physical exploration, without any linguistic supervision?
arXiv:2609.38927v1 Announce Type: cross Abstract: World models aim to learn representations of real-world environments and predict their future evolution. Recent object-centric world models have made...
The paper introduces the Geometry‑Native Autoencoder (GAE), a compact latent space that can be decoded into appearance, depth, camera parameters, and point maps, enabling 3D‑consistent world generation. By reparameterizing a geometry foundation model’s features, GAE replaces traditional appearance‑centric latents and improves visual quality and 3D coherence, achieving significant reductions in FVD and camera‑trajectory error on benchmark datasets. The work demonstrates that a geometry‑native latent space can serve as a shared interface between perception and generation models.
arXiv:2603. 19312v3 Announce Type: replace Abstract: Joint Embedding Predictive Architectures (JEPAs) offer a compelling framework for learning world models in compact latent spaces, yet existing methods remain fragile, relying on complex multi-term losses, exponential moving averages, pre-trained encoders, or auxiliary supervision to avoid representation collapse.
The paper investigates the Platonic Representation Hypothesis, which posits that more capable models converge toward shared representations. By distinguishing relational structure (which samples are related) from metric geometry (quantitative relations like distances), the authors develop a controlled $2 imes2$ framework to evaluate both aspects at local and global scales. Their findings show that relational structure consistently converges across vision‑language and video‑text models, while metric geometry converges much more weakly, a pattern that persists even when using a Riemannian metric approximation.
The paper introduces ATLAS, a training objective that preserves relational geometry in latent world models while calibrating the global latent distribution. By transferring normalized pairwise structure from an informative encoder to the planning latent and applying Wasserstein embedding matching, ATLAS improves goal‑reaching success on tasks such as PushT, TwoRoom, and OGBench‑Cube, especially on higher‑novelty episodes. Diagnostics show stronger novelty‑related structure, better marginal calibration, and lower multi‑step prediction error in the planning latent.
The paper proves the Platonic Representation Hypothesis by showing that Joint-Embedding Predictive Architectures (JEPAs) trained on different modalities converge to the same representation up to a rotation when their positive pairs are good views of the same scene. It demonstrates that next‑token prediction is a JEPA with bad views, explaining why language models only align with vision through large scale. Experiments on millions of captions and web images confirm a scaling law of alignment, revealing that good views require far fewer samples than next‑token prediction and that a 109M‑parameter text JEPA outperforms larger language models in aligning with image encoders, even enabling a simple rotation to produce an image captioner with high COCO CIDEr scores.
arXiv:2610.15429v1 Announce Type: cross Abstract: How do we tell a world model what information to preserve? We study this question through the lens of spectral representation learning. We introduce...
arXiv:2610.28074v1 Announce Type: new Abstract: As a prototypical latent world model, Joint-Embedding Predictive Architectures (JEPAs) provide strong semantic representations. However, dense patch re...
WALT introduces a method to align latent trajectories with pretrained driving world models, creating a compact generative trajectory space that preserves action-relevant semantics without altering the original model. The approach uses a dual-branch autoencoder to map raw waypoints into this latent space and transfers visual world knowledge into trajectory representations. Experiments on NAVSIM benchmarks show modest performance gains and a 30.5% reduction in planner FLOPs, indicating that maintaining world representations while extracting action-relevant information can improve trajectory planning efficiency.
JEPA-Bisim introduces a bisimulation encoder to joint-embedding predictive world models, ensuring that states with similar transition dynamics are mapped to nearby latent representations while suppressing irrelevant slow features such as background changes and distractors. The approach improves robustness on navigation (PointMaze) and manipulation (PushT) tasks under varied test-time visual conditions, achieving up to tenfold smaller latent spaces than DINO-WM. It remains effective across different pre-trained visual encoders, including DINOv2, SimDINOv2, and iBOT.