Demo-JEPA: Joint-Embedding Predictive Architecture for One-shot Cross-Embodiment Imitation

arXiv 2026 46.4 method

TLDR

Demo-JEPA uses a JEPA-based world model for one-shot cross-embodiment imitation by predicting future latent trajectories from visual demonstrations.

Reasoning

The paper presents a novel framework that decouples demonstration intent from embodiment-specific actions, leveraging a world model for latent trajectory prediction. Strengths include strong empirical results on both simulated and real-world tasks, while weaknesses may include limited discussion of failure cases or scalability to more complex embodiments.

Read-first score

Read-first score 46.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 29.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

80.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.80414777

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=experiment

Topical relevance 29%

41.4

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 310.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

generative world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Decoupling demonstration intent from embodiment-specific execution
Using JEPA-based world model to translate source visual demonstrations into target-compatible future latent trajectories
Avoiding action-level correspondence, requiring only visual demonstrations and target agent's own interaction experience
One-shot cross-embodiment imitation

Methodology

Demo-JEPA is built on a JEPA-based world model. It translates source visual demonstrations into target-compatible future latent trajectories in a shared predictive representation space. The target agent then uses these latent trajectories as subgoals and realizes them through planning under its own learned forward dynamics, requiring only visual demonstrations plus the target agent's own interaction experience.

Key Results

Experiments on RLBench and real-world manipulation tasks show that Demo-JEPA matches specialized in-domain planners and generalizes to unseen tasks and embodiment configurations where prior methods fail.

Resources

Paper