Latent Dynamics · 2026-10-06
EpicWorldModel: Exploration-driven Planning with Latent World Models
Bowen Feng, Julian Ost, May Mei, Anirudha Majumdar, Felix Heide
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
EpicWorldModel addresses ambiguity under partial observability in latent-space planning by training a stochastic JEPA world model. It uses a conditional flow-matching objective to generate multiple plausible future latent states, then estimates uncertainty via flow predictive variance (ensemble disagreement). This uncertainty is added to CEM planning to balance goal reaching with exploration; results show best or on-par performance and up to 22% higher success rate over LeWorldModel.
Why it matters
Deterministic JEPA world models can hallucinate incorrect futures when the goal-relevant content is occluded or absent from the conditioning history. EpicWorldModel learns a conditional distribution of future latent states and uses predictive variance as an exploration signal for planning, enabling agents to search uncertain regions more effectively under partial observability.
Method
- Train a stochastic JEPA: replace deterministic latent regression with conditional distribution prediction via conditional flow matching in Gaussian latent space, jointly learning encoder and stochastic predictor.
- Use flow predictive variance from multiple sampled futures (ensemble disagreement) as an exploration guidance; motivated as an upper bound proxy for predictive entropy and connected to expected information gain.
- Incorporate the uncertainty signal into CEM-based MPC planning to balance goal reaching and exploration of uncertain regions (e.g., occluded goals).
Abstract (from arXiv)
Latent world models based on Joint-Embedding Predictive Architecture (JEPA) are deterministic by design. While successful in fully observable scenarios, this paradigm breaks down when past observations and actions lead to multiple plausible future possibilities, e.g., due to occlusion. We introduce EpicWorldModel, a framework to train stochastic JEPAs for environments and tasks with inherent uncertainty under partially observability. We jointly train the EpicWorldModel predictor with its latent representation space to directly predict multiple potential future states using a flow-matching objective, when the goal-relevant scene content is absent from the conditioning history. We show that flow predictive variance, motivated by its relation to an upper bound on predictive entropy, serves as a useful exploration guidance for planning. By incorporating this uncertainty signal into Cross-Entropy Method (CEM)-based planning, our approach balances goal-reaching with exploration of uncertain regions where occluded goals are most likely to be located. We demonstrate the effectiveness of EpicWorldModel through a series of latent planning experiments with the best or on-par performance across tasks, showing up to 22% empirical improvement in success rate over LeWorldModel.
Related papers
- ATLAS: Aligned Transport of Latent Structure for Reliable World Model Planning
- Beyond a single latent space: a dual-latent world model for long-horizon planning
- SkipWorld: Latent Skip Planning with Distilled World Models for Efficient Visual Navigation
- How Long, Not How Close: A Learned Temporal Metric for Planning in Latent World Models
- Enhancing Latent World Models with Associative Memory
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-05 · Indexed by Pub-AI: 2026-10-06 UTC
EpicWorldModel: Exploration-driven Planning with Latent World Models
Preprint posted: 2026-10-05 · Original record ↗
- arxiv
- 2610.05996
Explicit source versions (1)
- v1 · 2026-10-05
- Preprint available
Source date: 2026-10-05 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Latent world models based on Joint-Embedding Predictive Architecture (JEPA) are deterministic by design. While successful in fully observable scenarios, this paradigm breaks down when past observations and actions lead to multiple plausible future possibilities, e.g., due to occlusion. We introduce EpicWorldModel, a framework to train stochastic JEPAs for environments and tasks with inherent uncertainty under partially observability. We jointly train the EpicWorldModel predictor with its latent representation space to directly predict multiple potential future states using a flow-matching objective, when the goal-relevant scene content is absent from the conditioning history. We show that flow predictive variance, motivated by its relation to an upper bound on predictive entropy, serves as a useful exploration guidance for planning. By incorporating this uncertainty signal into Cross-Entropy Method (CEM)-based planning, our approach balances goal-reaching with exploration of uncertain regions where occluded goals are most likely to be located. We demonstrate the effectiveness of EpicWorldModel through a series of latent planning experiments with the best or on-par performance across tasks, showing up to 22% empirical improvement in success rate over LeWorldModel.