arXiv:2605.25313v1 Announce Type: cross 
Abstract: World models for partially observed environments must imagine multiple compatible hidden futures and steer between them under counterfactual actions. Joint Embedding Predictive Architectures (JEPAs) do this in latent space, but a vector-valued latent has no internal structure for carrying the belief over hidden continuations through blind rollout. We introduce the Unitary World Model JEPA (UWM-JEPA), a JEPA world model with a density-matrix latent on a joint system-environment space and a learned unitary predictor. The construction preserves the joint-state spectrum exactly during rollout, so the predictor itself cannot dissipate the represented uncertainty.
  On a hidden-velocity indicator task requiring five-step forward simulation under a given action sequence with the target observation masked, UWM-JEPA reaches 0.77 accuracy and degrades monotonically as actions are perturbed; a parameter-matched LSTM-JEPA trained under the same counterfactual-target objective and action head collapses to majority-class accuracy (0.53) under every action condition. Under blind rollout, UWM-JEPA loses fewer than ten points of probe R^2 at short horizons while vector-latent baselines lose forty-one and sixty-eight; both nevertheless tie on a held-out context probe, locating the separation in the predictor rather than the encoder. Action sensitivity itself requires training against counterfactual rather than teacher-forced targets, a finding that applies beyond the unitary parameterisation. For JEPA world models to imagine under partial observability, latent geometry and predictor dynamics matter, not frozen context-encoding capacity alone.

تم تقديم نموذج العالم الموحد JEPA (UWM-JEPA) كنموذج تنبؤي جديد يعمل في فضاء الاعتقاد، مما يسمح بمحاكاة مستقبلات خفية متعددة في بيئات جزئية الملاحظة. يحسن هذا النموذج من قدرات الهياكل التنبؤية المدمجة المشتركة (JEPA) من خلال استخدام تمثيل كامن لمصفوفة الكثافة ومؤشر موحد متعلم، مما يضمن الحفاظ على عدم اليقين أثناء التنفيذ.

Se ha introducido el Modelo Mundial Unitario JEPA (UWM-JEPA) como un nuevo modelo predictivo que opera en el espacio de creencias, permitiendo la simulación de múltiples futuros ocultos en entornos parcialmente observados. Este modelo mejora las capacidades de las Arquitecturas Predictivas de Embedding Conjunto (JEPA) al utilizar una representación latente de matriz de densidad y un predictor unitario aprendido, asegurando que la incertidumbre se preserve durante el despliegue.

Le modèle Unitary World Model JEPA (UWM-JEPA) a été introduit comme un nouveau modèle prédictif qui opère dans l'espace de croyance, permettant la simulation de multiples futurs cachés dans des environnements partiellement observés. Ce modèle améliore les capacités des architectures prédictives d'embedding conjoint (JEPA) en utilisant une représentation latente de matrice de densité et un prédicteur unitaire appris, garantissant que l'incertitude est préservée lors du déploiement.

The Unitary World Model JEPA (UWM-JEPA) has been introduced as a novel predictive world model that operates in belief space, allowing for the simulation of multiple hidden futures in partially observed environments. This model enhances the capabilities of Joint Embedding Predictive Architectures (JEPAs) by utilizing a density-matrix latent representation and a learned unitary predictor, ensuring that uncertainty is preserved during rollout.