Saved in:
Bibliographic Details
Main Authors: Chen, Chang, Hamed, Hany, Baek, Doojin, Kang, Taegu, Noh, Samyeul, Bengio, Yoshua, Ahn, Sungjin
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2503.20102
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917082288881664
author Chen, Chang
Hamed, Hany
Baek, Doojin
Kang, Taegu
Noh, Samyeul
Bengio, Yoshua
Ahn, Sungjin
author_facet Chen, Chang
Hamed, Hany
Baek, Doojin
Kang, Taegu
Noh, Samyeul
Bengio, Yoshua
Ahn, Sungjin
contents Long-horizon planning is crucial in complex environments, but diffusion-based planners like Diffuser are limited by the trajectory lengths observed during training. This creates a dilemma: long trajectories are needed for effective planning, yet they degrade model performance. In this paper, we introduce this extendable long-horizon planning challenge and propose a two-phase solution. First, Progressive Trajectory Extension incrementally constructs longer trajectories through multi-round compositional stitching. Second, the Hierarchical Multiscale Diffuser enables efficient training and inference over long horizons by reasoning across temporal scales. To avoid the need for multiple separate models, we propose Adaptive Plan Pondering and the Recursive HM-Diffuser, which unify hierarchical planning within a single model. Experiments show our approach yields strong performance gains, advancing scalable and efficient decision-making over long-horizons.
format Preprint
id arxiv_https___arxiv_org_abs_2503_20102
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Extendable Planning via Multiscale Diffusion
Chen, Chang
Hamed, Hany
Baek, Doojin
Kang, Taegu
Noh, Samyeul
Bengio, Yoshua
Ahn, Sungjin
Machine Learning
Robotics
Long-horizon planning is crucial in complex environments, but diffusion-based planners like Diffuser are limited by the trajectory lengths observed during training. This creates a dilemma: long trajectories are needed for effective planning, yet they degrade model performance. In this paper, we introduce this extendable long-horizon planning challenge and propose a two-phase solution. First, Progressive Trajectory Extension incrementally constructs longer trajectories through multi-round compositional stitching. Second, the Hierarchical Multiscale Diffuser enables efficient training and inference over long horizons by reasoning across temporal scales. To avoid the need for multiple separate models, we propose Adaptive Plan Pondering and the Recursive HM-Diffuser, which unify hierarchical planning within a single model. Experiments show our approach yields strong performance gains, advancing scalable and efficient decision-making over long-horizons.
title Extendable Planning via Multiscale Diffusion
topic Machine Learning
Robotics
url https://arxiv.org/abs/2503.20102