Newest papers

Rethinking Video Autoencoders for Interactive World Models

Adrien Ramanana Rahary, Yutong Bai, Nicolas Dufour, Patrick Pérez, David Picard · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30129 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Los modelos de cámaras del mundo de video sporg predicen frames futuros de frames y acciones pasadas. Sirven como simuladores en los que los agentes entrenan, planean o se evalúan, y que los humanos pueden controlar de forma interactiva. Generan en el espacio latente de un codificador de vídeo en movimiento, cuya compresión establece su coste por frames y cuánto tiempo espera cada acción para tener efecto. Sin embargo, los modelos del mundo a menudo reutilizan los codificadores de los generadores de texto a vídeo, construidos para la generación fuera de línea sin agentes de tiempo efectivo y, sin sus datos de entrenamiento y código, no son retrainables en otra compresión. Por lo tanto, preguntamos cómo, dónde y cuánto debe comprimirse un modelo del mundo. Construimos CADRE, una familia abierta de cod

Machine translation (Meta M2M100); the English original follows.

Abstract

Video world models predict future frames from past frames and actions. They serve as simulators in which agents train, plan or are evaluated, and that humans can control interactively. They generate in the latent space of a video autoencoder, whose compression sets their cost per frame and how long each action waits to take effect. Yet world models often reuse the autoencoders of text-to-video generators, built for offline generation with no agent waiting and, without their training data and code, not retrainable at another compression. We therefore ask how, where and how much a world model should compress. We build CADRE, an open family of causal diffusion representation autoencoders spanning multiple spatio-temporal compressions. Its members differ only in a small bottleneck on frozen features and share a diffusion decoder self-distilled to one or a few steps. We train the same world model on each autoencoder in two domains, manipulation and navigation, with two environments each. CADRE forms most of the latency-quality frontier, especially with a moving camera. Our study also leads to four findings. At strong compression, a diffusion decoder is sharper than perceptual regression. In robot manipulation, temporal compression is more effective in the autoencoder than in the world model. Manipulation and navigation call for opposite compressions: temporal compression costs little when the camera is still, and spatial compression when it moves. Surprisingly, with a still camera, coarser spatial tokens hurt motion far more than image quality. Code, models and a live demo are at https://kyutai.org/blog/2026-10-21-cadre

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Autoencoder, Artificial intelligence, Data compression, Frame (networking), Bottleneck

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange