Spectral-Target Physical Latent Structuring for JEPA-Style World Models

Penghao Zhu, Salvatore Penachio, Kaustav Mukherjee, Aneesh Jonelagadda

arXiv:2609.04264 · 2026-09-07 공개 · arXiv · PDF

latent-world-models planning-success jeppa leworldmodel latent-space-structuring low-data-regimes sigreg fourier-auxiliary-head

Abstract

Latent world models have become increasingly popular as a method to predict and plan in latent space rather than pixel space. Recent architectures, such as LeWorldModel (LeWM), jointly train the encoder and predictor using regularization techniques like SIGReg to prevent representation collapse. Even with such regularization preventing representation collapse, we identify a new world model failure mode of \textit{physical representation laziness}, particularly noted in highly dynamic environments. For these lazy cases, the learned latent states do not collapse but nonetheless fail to represent key physical properties, causing ubiquitous downstream planning failure. To resolve this issue, we propose training-time auxiliary supervision with a lightweight "Fourier auxiliary head", which enforces physically-informed structuring of the latent space with no additional inference-time cost and can be generalized to any environment. Experimentally, we show that the auxiliary head substantially improves planning success rates in dynamic environments where the baseline LeWM exhibits physical representation laziness. It also leads to modest improvements in other environments, even when the baseline does not exhibit physical representation laziness. We further observe superior planning performance being accompanied by higher latent space correlations with key physical properties, indicating both the ability of our method to physically structure latent states and the potential planning-side benefit to the learned representation being physically structured. We also see in low-data regimes, auxiliary supervision is particularly impactful in increasing success rate. These findings support the use of our Fourier auxiliary head method to improve both overall success rate and data efficiency, while avoiding representation laziness in latent world models.

한국어 요약

한 줄 요약

JEPA 기반 월드 모델에서 물리적 정보를 효과적으로 인코딩하기 위해 Fourier auxiliary head를 도입하여 planning 성능과 데이터 효율성을 향상시킨다.

핵심 기여도

핵심 아이디어

JEPA 기반 월드 모델은 latent space에서 예측과 계획을 수행하지만, 기존의 SIGReg는 representation collapse를 방지하지만, 핵심 물리적 정보를 생략하는 **representation laziness**를 막지 못한다. 이는 downstream planning 성능에 부정적인 영향을 미친다. 이를 해결하기 위해, 저자는 **Fourier auxiliary head**를 제안한다. 이 모듈은 training 시에만 사용되며, task-relevant object의 위치나 bounding box를 기반으로 latent state에서 **spectral features**를 복원하도록 학습한다. 이는 latent space가 물리적 정보를 반영하도록 유도하며, planning 성능을 향상시킨다. 특히, 이 모듈은 **inference 시 제거 가능**하며, 추가적인 계산 비용이 발생하지 않는다.

기술적 접근법

주요 결과

의의 및 한계

실용적 활용