mlsanityData to 13 Oct 2026

JEPA for RL: Investigating Joint-Embedding Predictive Architectures for Reinforcement Learning

arXiv 2504.16591 ↗First posted 23 Apr 2025cs.CVcited by 4

On the map: Deep reinforcement learning methods → RL for robot locomotion and manipulation

Abstract (arXiv)

Joint-Embedding Predictive Architectures (JEPA) have recently become popular as promising architectures for self-supervised learning. Vision transformers have been trained using JEPA to produce embeddings from images and videos, which have been shown to be highly suitable for downstream tasks like classification and segmentation. In this paper, we show how to adapt the JEPA architecture to reinforcement learning from images. We discuss model collapse, show how to prevent it, and provide exemplary data on the classical Cart Pole task.

Key facts. Cited by 4 papers in the map (data through 13 Oct 2026). Placed in RL for robot locomotion and manipulation, one of 15 directions in Deep reinforcement learning methods. Builds on 6 of its 13 references that are in the map. First posted on arXiv 23 Apr 2025.

Builds on

6 of 13 references.

References · 6

  1. 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,730
  2. 2018World Modelscited by 1,481
  3. 2021VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learningcited by 1,033
  4. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 1,018
  5. 2020CURL: Contrastive Unsupervised Representations for Reinforcement Learningcited by 747
  6. 2018Understanding disentangling in β-VAEcited by 334

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2025: 2; 2026: 2.

4 papers.

  1. 2025LLM-JEPA: Large Language Models Meet Joint Embedding Predictive Architecturescited by 38
  2. 2026Semantic Tube Prediction: Beating LLM Data Efficiency with JEPAcited by 6
  3. 2025Why and How Auxiliary Tasks Improve JEPA Representationscited by 5
  4. 2026The JEPA Predictor: A Transferable Operator for Occluded Feature Completioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026MA-JEPA: Joint-Embedding World Models for Multi-Agent Reinforcement Learning
  2. 2026VJEPA: Variational Joint Embedding Predictive Architectures as Probabilistic World Models
  3. 2026A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures
  4. 2022Joint Embedding Predictive Architectures Focus on Slow Features
  5. 2025Value-guided action planning with JEPA world models
  6. 2025ACT-JEPA: Novel Joint-Embedding Predictive Architecture for Efficient Policy Representation Learning
  7. 2026LeWAM: End-to-End World and Action Modeling with JEPAs
  8. 2026Hamiltonian JEPA: Action-Conditioned World Models with an Inherited Control State