Horizon Imagination: Efficient On-Policy Rollout in Diffusion World Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Cohen, Lior, Nabati, Ofir, Wang, Kaixin, Kumar, Navdeep, Mannor, Shie
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866912910334230528
author Cohen, Lior
Nabati, Ofir
Wang, Kaixin
Kumar, Navdeep
Mannor, Shie
author_facet Cohen, Lior
Nabati, Ofir
Wang, Kaixin
Kumar, Navdeep
Mannor, Shie
contents We study diffusion-based world models for reinforcement learning, which offer high generative fidelity but face critical efficiency challenges in control. Current methods either require heavyweight models at inference or rely on highly sequential imagination, both of which impose prohibitive computational costs. We propose Horizon Imagination (HI), an on-policy imagination process for discrete stochastic policies that denoises multiple future observations in parallel. HI incorporates a stabilization mechanism and a novel sampling schedule that decouples the denoising budget from the effective horizon over which denoising is applied while also supporting sub-frame budgets. Experiments on Atari 100K and Craftium show that our approach maintains control performance with a sub-frame budget of half the denoising steps and achieves superior generation quality under varied schedules. Code is available at https://github.com/leor-c/horizon-imagination.
format Preprint
id arxiv_https___arxiv_org_abs_2602_08032
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Horizon Imagination: Efficient On-Policy Rollout in Diffusion World Models
Cohen, Lior
Nabati, Ofir
Wang, Kaixin
Kumar, Navdeep
Mannor, Shie
Machine Learning
We study diffusion-based world models for reinforcement learning, which offer high generative fidelity but face critical efficiency challenges in control. Current methods either require heavyweight models at inference or rely on highly sequential imagination, both of which impose prohibitive computational costs. We propose Horizon Imagination (HI), an on-policy imagination process for discrete stochastic policies that denoises multiple future observations in parallel. HI incorporates a stabilization mechanism and a novel sampling schedule that decouples the denoising budget from the effective horizon over which denoising is applied while also supporting sub-frame budgets. Experiments on Atari 100K and Craftium show that our approach maintains control performance with a sub-frame budget of half the denoising steps and achieves superior generation quality under varied schedules. Code is available at https://github.com/leor-c/horizon-imagination.
title Horizon Imagination: Efficient On-Policy Rollout in Diffusion World Models
topic Machine Learning
url https://arxiv.org/abs/2602.08032