ProWorld: Progress-Aware Hyperbolic World Models for Long-Horizon Visual Goal Reaching

2026-08-03

AI Summary

This work addresses the limitation of existing visual world models in long-horizon tasks, which rely solely on local state prediction and often fail to maintain consistent progress toward a goal, resulting in trajectory deviations or ambiguous advancement. To overcome this, the authors propose a goal-conditioned hyperbolic visual world model that, for the first time, integrates ordinal progress relationships with hyperbolic geometry. By leveraging hyperbolic entailment learning and a future state discrimination mechanism, the model explicitly encodes directional and hierarchical structure of state progression in latent space. A progress-aware planning objective is further introduced to guide decision-making. Built upon JEPA-style modeling and hyperbolic embeddings, the method achieves an average success rate improvement of 9.67 percentage points over LeWM across four visual goal-reaching benchmarks, demonstrating significantly enhanced long-horizon planning capability.

Abstract

JEPA-style visual world models offer an effective paradigm for visual goal planning by predicting future latent representations. Existing methods typically learn local transition consistency through next-step representation prediction. However, in long-horizon tasks, accurate local prediction alone need not ensure sustained progress toward the goal. First, multi-step rollouts can remain locally plausible while drifting away from goal-relevant trajectories. Second, locally similar future states can correspond to substantially different long-term progress, making them difficult to distinguish in a latent space optimized mainly for local consistency. To address these challenges, we introduce goal-conditioned progress order, a relative ordering of states according to how they advance toward a given goal. This order exhibits an asymmetric, coarse-to-fine structure: early states retain broader future possibilities, while later states concentrate on more specific goal-relevant regions. Such a structure is well suited to hyperbolic geometry. Motivated by this observation, we propose ProWorld, a progress-aware hyperbolic visual world model. ProWorld leverages goal-conditioned progress order to organize visual latent-space dynamics, maintains directional progress within trajectories via hyperbolic entailment learning, and mitigates progress ambiguity among locally similar future states via hyperbolic future discrimination. Furthermore, we design a progress-aware planning objective that scores candidate rollouts by jointly considering proximity to the goal and sustained progress across intermediate states. Experiments on four visual goal-reaching tasks demonstrate that ProWorld achieves an average absolute success-rate gain of 9.67 over LeWM. The code will be released after the paper is accepted.

Problem

• long-horizon visual goal reaching
• visual world models
• progress awareness
• latent space ambiguity
• goal-directed planning

Innovation

• hyperbolic geometry
• goal-conditioned progress order
• visual world model
• long-horizon planning
• progress-aware representation