Knowledge Diversion for Efficient Morphology Control and Policy Transfer

TL;DR

DivMorph introduces a modular Transformer-based framework that disentangles shared and task-specific knowledge, enabling scalable, efficient morphology control and transfer across heterogeneous agents with significant improvements in sample efficiency and model size.

Contribution

It proposes a novel knowledge diversion method that factorizes Transformer weights and uses dynamic gating for scalable, transferable morphology control.

Findings

3x improvement in sample efficiency for cross-task transfer

17x reduction in model size for deployment

State-of-the-art performance in morphology control

Abstract

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.