mlsanityData to 7 Oct 2026

Next Embedding Prediction Makes World Models Stronger

arXiv 2603.02765 ↗First posted 3 Mar 2026cs.LG cs.AIcited by 4

On the map: World models → Latent dynamics models and model-based RL

Abstract (arXiv)

Capturing temporal dependencies is critical for model-based reinforcement learning (MBRL) in partially observable, high-dimensional domains. We introduce NE-Dreamer, a decoder-free MBRL agent that leverages a temporal transformer to predict next-step encoder embeddings from latent state sequences, directly optimizing temporal predictive alignment in representation space. This approach enables NE-Dreamer to learn coherent, predictive state representations without reconstruction losses or auxiliary supervision. On the DeepMind Control Suite, NE-Dreamer matches or exceeds the performance of DreamerV3 and leading decoder-free agents. On a challenging subset of DMLab tasks involving memory and spatial reasoning, NE-Dreamer achieves substantial gains. These results establish next-embedding prediction with temporal transformers as an effective, scalable framework for MBRL in complex, partially observable environments.

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Placed in Latent dynamics models and model-based RL, one of 11 directions in World models. Builds on 23 of its 30 references that are in the map. First posted on arXiv 3 Mar 2026.

Builds on

23 of 30 references.

Closest ancestors · 8

Same direction, newest first

  1. 2024MuDreamer: Learning Predictive World Models without Reconstruction23 May
  2. 2023TD-MPC2: Scalable, Robust World Models for Continuous Control25 Oct
  3. 2022DreamingV2: Reinforcement Learning with Discrete World Models without Reconstruction1 Mar
  4. 2021DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations27 Oct
  5. 2020Mastering Atari with Discrete World Models5 Oct
  6. 2020Dreaming: Model-based Reinforcement Learning by Latent Imagination without Reconstruction29 Jul
  7. 2019Dream to Control: Learning Behaviors by Latent Imagination3 Dec
  8. 2018World Models27 Mar

Other references · 15

  1. 2018Representation Learning with Contrastive Predictive Codingcited by 7,662
  2. 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,712
  3. 2020Exploring Simple Siamese Representation Learningcited by 2,853
  4. 2021Barlow Twins: Self-Supervised Learning via Redundancy Reductioncited by 1,818
  5. 2019Mastering Atari, Go, Chess and Shogi by Planning with a Learned Modelcited by 1,327
  6. 2018Learning Latent Dynamics for Planning from Pixelscited by 1,221
  7. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 982
  8. 2022data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Languagecited by 718
  9. 2020Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixelscited by 573
  10. 2020Reinforcement Learning with Augmented Datacited by 452
  11. 2022Temporal Difference Learning for Model Predictive Controlcited by 375
  12. 2020dm_control: Software and Tasks for Continuous Controlcited by 368
3 more
  1. 2020Data-Efficient Reinforcement Learning with Self-Predictive Representationscited by 309
  2. 2021The Distracting Control Suite -- A Challenging Benchmark for Reinforcement Learning from Pixelscited by 97
  3. 2025Next-Embedding Prediction Makes Strong Vision Learnerscited by 21

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026Looped World Modelscited by 2
  2. 2026Rethinking Point Clouds as Sequences: A Causal Next-Token Predictive Learning Frameworkcited by 1
  3. 2026Pixel-level Scene Understanding in One Token: Visual States Need What-is-Where Compositioncited by 0
  4. 2026Listening Forward: Next Patch Embedding Prediction Enables Scalable Audio Learnerscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Next-Latent Prediction Transformers Learn Compact World Models
  2. 2026Dreamer-CDP: Improving Reconstruction-free World Models Via Continuous Deterministic Representation Prediction
  3. 2026Temporal-Distance JEPA: Plan-Aware Representation Learning for Latent World Model Predictive Control
  4. 2025TD-JEPA: Latent-predictive Representations for Zero-Shot Reinforcement Learning
  5. 2026R2-Dreamer: Redundancy-Reduced World Models without Decoders or Augmentation
  6. 2021Temporal Predictive Coding For Model-Based Planning In Latent Space
  7. 2026MA-JEPA: Joint-Embedding World Models for Multi-Agent Reinforcement Learning
  8. 2023Facing Off World Model Backbones: RNNs, Transformers, and S4