Causal-JEPA: Learning World Models through Object-Level Latent Interventions

TL;DR

C-JEPA introduces an object-centric world model with object-level masking that enhances relational understanding, improves counterfactual reasoning, and enables more efficient planning in control tasks.

Contribution

It extends masked joint embedding prediction to object-centric representations, inducing causal latent interventions and improving interaction reasoning.

Findings

20% improvement in counterfactual reasoning for visual question answering

Achieves comparable control performance using only 1% of latent features

Formal analysis shows object-level masking induces causal inductive bias

Abstract

World models require robust relational understanding to support prediction, reasoning, and control. While object-centric representations provide a useful abstraction, they are not sufficient to capture interaction-dependent dynamics. We therefore propose C-JEPA, a simple and flexible object-centric world model that extends masked joint embedding prediction from image patches to object-centric representations. By applying object-level masking that requires an object's state to be inferred from other objects, C-JEPA induces latent interventions with counterfactual-like effects and prevents shortcut solutions, making interaction reasoning essential. Empirically, C-JEPA leads to consistent gains in visual question answering, with an absolute improvement of about 20\% in counterfactual reasoning compared to the same architecture without object-level masking. On agent control tasks, C-JEPA enables substantially more efficient planning by using only 1\% of the total latent input features required by patch-based world models, while achieving comparable performance. Finally, we provide a formal analysis demonstrating that object-level masking induces a causal inductive bias via latent interventions. Our code is available at https://github.com/galilai-group/cjepa.