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

📰 ArXiv cs.AI

MSG improves sample efficiency in robotic manipulation with multi-stream generative policies

advanced Published 1 Apr 2026

Action Steps

Train multiple object-centric policies
Combine policies at inference time using MSG framework
Improve generalization and sample efficiency in robotic manipulation tasks
Evaluate MSG on various robotic tasks to demonstrate its effectiveness

Who Needs to Know This

Robotics engineers and AI researchers benefit from MSG as it enhances generalization and sample efficiency in robotic manipulation tasks, allowing for more effective policy learning

Key Insight

💡 Combining multiple object-centric policies at inference time can improve generalization and sample efficiency in robotic manipulation

Key Takeaways

MSG improves sample efficiency in robotic manipulation with multi-stream generative policies

Full Article

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

Abstract:
arXiv:2509.24956v2 Announce Type: replace-cross Abstract: 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 efficienc

Read full paper → ← Back to Reads