JEPA-Bisim: Learning Robust Visual Representations for Planning with Joint-Embedding Predictive World Models

arXiv:2602.18639 · cs.LG, math.OC · Submitted 2026-02-20 · Read on arXiv

cs.LG, math.OC

Submitted: 2026-02-20

Updated: 2026-10-01

Comments: This version refines the text and provides additional numerical validation

Code: https://github.com/jd-anderson/dino_bsmpc

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Robustness — 3× in this paper · explained in 5 episodes
• World Model — 3× in this paper · explained in 7 episodes
• Latent Space — 2× in this paper · explained in 7 episodes
• Benchmark — 1× in this paper · explained in 2 episodes

Sources

• LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
• VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• D4RL: Datasets for Deep Data-Driven Reinforcement Learning
• World Models
• From CLIP to DINO: Visual Encoders Shout in Multi-modal Large Language Models
• Performance Asymmetry in Model-Based Reinforcement Learning
• LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels
• DINOv2: Learning Robust Visual Features without Supervision
• Foundation Policies with Hilbert Representations
• Closing the Train-Test Gap in World Models for Gradient-Based Planning
• Joint Embedding Predictive Architectures Focus on Slow Features
• Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models
• Learning Latent Dynamic Robust Representations for World Models
• A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures
• Temporal Straightening for Latent Planning
• Simplifying DINO via Coding Rate Regularization
• iBOT: Image BERT Pre-Training with Online Tokenizer
• Variance-Covariance Regularization Improves Representation Learning

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks