Saved in:
Bibliographic Details
Main Authors: Pu, Yuan, Niu, Yazhe, Tang, Jia, Xiong, Junyu, Hu, Shuai, Li, Hongsheng
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.07945
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915838860197888
author Pu, Yuan
Niu, Yazhe
Tang, Jia
Xiong, Junyu
Hu, Shuai
Li, Hongsheng
author_facet Pu, Yuan
Niu, Yazhe
Tang, Jia
Xiong, Junyu
Hu, Shuai
Li, Hongsheng
contents In heterogeneous multi-task decision-making, tasks not only exhibit diverse observation and action spaces but also vary substantially in their underlying complexities. While conventional multi-task world models like UniZero excel in single-task settings, we find that when handling a broad and diverse suite of tasks, gradient conflicts and the loss of model plasticity often constrain their sample efficiency. In this work, we address these challenges from two complementary perspectives: the single learning iteration and the overall learning process. First, to mitigate the gradient conflicts, we systematically investigate key architectural designs for extending UniZero. Our investigation identifies a Mixture-of-Experts (MoE) architecture as the most effective approach. We demonstrate, both theoretically and empirically, that this architecture alleviates gradient conflicts by routing task-specific representations to specialized sub-networks. This finding leads to our proposed model, \textit{ScaleZero}. Second, to dynamically allocate model capacity throughout the learning process, we introduce an online Dynamic Parameter Scaling (DPS) strategy. This strategy progressively integrates LoRA adapters in response to task-specific progress, enabling adaptive knowledge retention and parameter expansion. Evaluations on a diverse set of standard benchmarks (Atari, DMC, Jericho) demonstrate that ScaleZero, utilizing solely online reinforcement learning with one model, performs on par with specialized single-task agents. With the DPS strategy, it remains competitive while using just 71.5% of the environment interactions. These findings underscore the potential of ScaleZero for effective multi-task planning. Our code is available at \textcolor{magenta}{https://github.com/opendilab/LightZero}.
format Preprint
id arxiv_https___arxiv_org_abs_2509_07945
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle One Model for All Tasks: Leveraging Efficient World Models in Multi-Task Planning
Pu, Yuan
Niu, Yazhe
Tang, Jia
Xiong, Junyu
Hu, Shuai
Li, Hongsheng
Machine Learning
In heterogeneous multi-task decision-making, tasks not only exhibit diverse observation and action spaces but also vary substantially in their underlying complexities. While conventional multi-task world models like UniZero excel in single-task settings, we find that when handling a broad and diverse suite of tasks, gradient conflicts and the loss of model plasticity often constrain their sample efficiency. In this work, we address these challenges from two complementary perspectives: the single learning iteration and the overall learning process. First, to mitigate the gradient conflicts, we systematically investigate key architectural designs for extending UniZero. Our investigation identifies a Mixture-of-Experts (MoE) architecture as the most effective approach. We demonstrate, both theoretically and empirically, that this architecture alleviates gradient conflicts by routing task-specific representations to specialized sub-networks. This finding leads to our proposed model, \textit{ScaleZero}. Second, to dynamically allocate model capacity throughout the learning process, we introduce an online Dynamic Parameter Scaling (DPS) strategy. This strategy progressively integrates LoRA adapters in response to task-specific progress, enabling adaptive knowledge retention and parameter expansion. Evaluations on a diverse set of standard benchmarks (Atari, DMC, Jericho) demonstrate that ScaleZero, utilizing solely online reinforcement learning with one model, performs on par with specialized single-task agents. With the DPS strategy, it remains competitive while using just 71.5% of the environment interactions. These findings underscore the potential of ScaleZero for effective multi-task planning. Our code is available at \textcolor{magenta}{https://github.com/opendilab/LightZero}.
title One Model for All Tasks: Leveraging Efficient World Models in Multi-Task Planning
topic Machine Learning
url https://arxiv.org/abs/2509.07945