The paper investigates why latent world models, despite accurate latent predictions, can perform poorly in downstream planning. It introduces the concept of action-identifiability and formalizes it via Bayes inverse risk, showing that self-decodable predicted transitions may encode domain‑specific action relationships that do not transfer to real environments. Using the Action‑Consistency Transfer Matrix (ATM), the authors demonstrate that true‑transition action‑identifiability correlates strongly with planning success across several benchmarks, and that the ATM can diagnose cross‑domain inconsistencies and aid lightweight model screening.
arXiv:2606. 09028v1 Announce Type: cross Abstract: Latent world models are increasingly used for control and goal-conditioned planning, yet assessing whether their learned representations are useful for planning usually requires slow, planner-coupled simulator evaluation with CEM or similar planners.
AD-WM is a new action‑discriminative joint‑embedding world model designed for counterfactual model predictive control. It augments residual latent dynamics with action‑recovery regularization based on inverse dynamics and conditional mutual information, while discarding auxiliary heads at test time so that MPC remains unchanged. Experiments on OGBench‑Cube and other simulation environments show substantial gains in hard‑start success and mean success, and zero‑shot transfer to a Franka robot improves pick‑and‑place success from 42.2% to 71.1%.
The paper introduces the Dual-Latent World Model (Dual-WM), which separates local execution and long-range planning into distinct latent spaces and dynamics models. A new learning method, Long-Horizon Representation Learning with Weighted Rollout (LoRe), supervises predictions at both levels using exponential horizon weights. Experiments on five goal-conditioned visual control tasks show that Dual-WM improves success rates over strong baselines, especially at longer horizons.
The paper presents an end‑to‑end JEPA world model that enhances latent prediction with inverse dynamics and state alignment to improve goal‑conditioned robotic planning. By preventing latent collapse and grounding representations in physical configuration, the model achieves top success rates on tasks such as TwoRoom, PushT, and OGBench‑Cube, outperforming the baseline LeWorldModel. Ablation studies confirm that state alignment consistently boosts planning success over inverse dynamics alone across all four benchmark tasks.
arXiv:2607. 17973v1 Announce Type: new Abstract: Latent world models have emerged as a powerful planning paradigm by learning action-conditioned predictive dynamics and using them as internal simulators to imagine and evaluate candidate action sequences.
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 presents an end‑to‑end JEPA world model that enhances latent prediction with inverse dynamics and state alignment to better support goal‑conditioned robotic planning. By incorporating inverse dynamics, the model prevents latent collapse and encodes action information, while state alignment ties consecutive latent states to their physical configurations and motions. Experiments on four benchmark tasks show the model achieves top success rates on TwoRoom, PushT, and OGBench‑Cube, and its state alignment consistently improves planning performance over inverse dynamics alone.
arXiv:2604. 03208v2 Announce Type: replace Abstract: World models are a promising path to zero-shot embodied control through planning.
The paper introduces Retrospective World Modeling, a new paradigm for vision‑language‑model (VLM) agents that allows them to reason backward by estimating which action most likely caused a state transition. It proposes the Self‑Consistency Reward (SCR), an intrinsic signal that measures how well a policy action aligns with this retrospective explanation, providing dense transition‑level feedback. Experiments demonstrate that incorporating SCR improves policy robustness and generalization compared to purely prospective world‑modeling approaches.
arXiv:2606. 27806v3 Announce Type: replace Abstract: Language agents plan by generating not only actions but also implicit predictions of how the world will change.
FlexiWorld is a JEPA-based latent world model that learns variable‑length action chunks across multiple time scales for goal‑directed planning. It jointly trains a causal action encoder and an autoregressive actor, using mixed‑span goal supervision and Student Forcing to reduce exposure bias. In experiments on four benchmarks, FlexiWorld with the Actor‑Residual Cross‑Entropy Method (ARCEM) achieves higher mean success rates than the strongest baseline and supports flexible planning chunk lengths without retraining.
arXiv:2608.24855v1 Announce Type: new Abstract: Latent world models are inherently strong encoders that transform image pixel to latent embedding, yet existing world models still rely on online traje...