MSG: Multi-Stream Generative Policies for Sample-Efficient Robotic Manipulation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: von Hartz, Jan Ole, Schweizer, Lukas, Boedecker, Joschka, Valada, Abhinav
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917374021599232
author von Hartz, Jan Ole
Schweizer, Lukas
Boedecker, Joschka
Valada, Abhinav
author_facet von Hartz, Jan Ole
Schweizer, Lukas
Boedecker, Joschka
Valada, Abhinav
contents Generative robot policies such as Flow Matching offer flexible, multi-modal policy learning but are sample-inefficient. Although object-centric policies improve sample efficiency, it does not resolve this limitation. In this work, we propose Multi-Stream Generative Policy (MSG), an inference-time composition framework that trains multiple object-centric policies and combines them at inference to improve generalization and sample efficiency. MSG is model-agnostic and inference-only, hence widely applicable to various generative policies and training paradigms. We perform extensive experiments both in simulation and on a real robot, demonstrating that our approach learns high-quality generative policies from as few as five demonstrations, resulting in a 95% reduction in demonstrations, and improves policy performance by 89 percent compared to single-stream approaches. Furthermore, we present comprehensive ablation studies on various composition strategies and provide practical recommendations for deployment. Finally, MSG enables zero-shot object instance transfer. We make our code publicly available at https://msg.cs.uni-freiburg.de.
format Preprint
id arxiv_https___arxiv_org_abs_2509_24956
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MSG: Multi-Stream Generative Policies for Sample-Efficient Robotic Manipulation
von Hartz, Jan Ole
Schweizer, Lukas
Boedecker, Joschka
Valada, Abhinav
Robotics
Artificial Intelligence
Machine Learning
Generative robot policies such as Flow Matching offer flexible, multi-modal policy learning but are sample-inefficient. Although object-centric policies improve sample efficiency, it does not resolve this limitation. In this work, we propose Multi-Stream Generative Policy (MSG), an inference-time composition framework that trains multiple object-centric policies and combines them at inference to improve generalization and sample efficiency. MSG is model-agnostic and inference-only, hence widely applicable to various generative policies and training paradigms. We perform extensive experiments both in simulation and on a real robot, demonstrating that our approach learns high-quality generative policies from as few as five demonstrations, resulting in a 95% reduction in demonstrations, and improves policy performance by 89 percent compared to single-stream approaches. Furthermore, we present comprehensive ablation studies on various composition strategies and provide practical recommendations for deployment. Finally, MSG enables zero-shot object instance transfer. We make our code publicly available at https://msg.cs.uni-freiburg.de.
title MSG: Multi-Stream Generative Policies for Sample-Efficient Robotic Manipulation
topic Robotics
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2509.24956