arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2512.09796cs.LG
收藏

知识分流用于高效形态控制与策略迁移

Knowledge Diversion for Efficient Morphology Control and Policy Transfer

School of Computer Science and Engineering, Southeast University, Nanjing, China(计算机科学与工程学院,东南大学,南京,中国);Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China(新一代人工智能技术及其交叉应用重点实验室(东南大学),教育部,中国)

Fu Feng, Ruixiao Shi, Yucheng Xie, Jianlu Shen, Jing Wang, Xin Geng

AI总结:

DivMorph通过知识分流实现高效形态控制与策略迁移,提升跨任务泛化能力与部署效率。

摘要:

通用形态控制旨在学习一个能够跨异质智能体形态泛化的大规模策略,基于Transformer的控制器已成为流行选择。然而,此类架构导致计算成本高昂,造成部署负担重,现有方法在跨任务泛化方面有限,需要为每个新任务从头开始训练。为此,我们提出了DivMorph,一种模块化训练范式,利用知识分流来学习可分解的控制器。DivMorph通过SVD将随机初始化的Transformer权重分解为因子单元,在训练前进行分解,并利用动态软门控根据任务和形态嵌入来调节这些单元,将它们分为共享的learngenes和形态和任务特定的tailors,从而实现知识解耦。通过选择性激活相关组件,DivMorph实现了可扩展且高效的策略部署,同时支持有效策略迁移至新任务。大量实验表明,DivMorph在跨任务迁移中实现了样本效率的3倍提升,在单智能体部署中实现了模型大小的17倍减少。

英文摘要:

Universal morphology control aims to learn a universal policy that generalizes across heterogeneous agent morphologies, with Transformer-based controllers emerging as a popular choice. However, such architectures incur substantial computational costs, resulting in high deployment overhead, and existing methods exhibit limited cross-task generalization, necessitating training from scratch for each new task. To this end, we propose \textbf{DivMorph}, a modular training paradigm that leverages knowledge diversion to learn decomposable controllers. DivMorph factorizes randomly initialized Transformer weights into factor units via SVD prior to training and employs dynamic soft gating to modulate these units based on task and morphology embeddings, separating them into shared \textit{learngenes} and morphology- and task-specific \textit{tailors}, thereby achieving knowledge disentanglement. By selectively activating relevant components, DivMorph enables scalable and efficient policy deployment while supporting effective policy transfer to novel tasks. Extensive experiments demonstrate that DivMorph achieves state-of-the-art performance, achieving a 3$\times$ improvement in sample efficiency over direct finetuning for cross-task transfer and a 17$\times$ reduction in model size for single-agent deployment.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑