ZICQ
中 Log in / Sign up
Newsroom Agentic #Shanghai AI Lab #InternW0 #Physical World Model #Robotic Control #Multimodal AI

Shanghai AI Lab Releases InternW0: A Groundbreaking Physical World Model for Efficient Human-Machine Interaction

Avatar of Mr.Xu

By Mr.Xu Compiled & Reviewed by Editorial

Published: · 14 views

中文阅读 (Chinese) English Version
Co-reading Select any sentence in the article to highlight, annotate, or copy a quote card. Join discussion ↓

Summary:Shanghai AI Laboratory introduces InternW0, the first instantiation of its InternW physical world model series. Built around omnimodal interfaces, asynchronous multi-frequency processing, and local physical modeling under partial observations and external influences, InternW0 excels in future visual dynamics prediction and continuous robot control. Its asymmetric video-action architecture with flow matching, combined with layer-wise K/V cache reuse, enables efficient processing and adaptation to


Key Breakthroughs

  • Omnimodal Interfaces and Asynchronous Multi-Frequency Processing: InternW0 leverages omnimodal interfaces and asynchronous multi-frequency processing to efficiently handle multi-source data in complex environments.
  • Asymmetric Video-Action Architecture: The model employs a high-capacity video expert for long-horizon predictive context and a lightweight action expert for rapid response, enabling efficient human-machine interaction.
  • Layer-wise K/V Cache Reuse: Instead of regenerating the future for every action update, InternW0 reuses layer-wise K/V cache and adapts it to new states through observation-conditioned context routing, significantly reducing computational costs.
  • Domain-Specific Interfaces and Soft Prompts: Supporting heterogeneous embodiments and incorporating force and tactile signals through contact-aware post-training, InternW0 excels in complex manipulation tasks.

Technical Highlights

  1. Efficient Prediction and Control: The innovative video-action architecture allows InternW0 to perform exceptionally well in long-horizon prediction and real-time control tasks.
  2. Resource Optimization: The layer-wise K/V cache reuse mechanism significantly lowers computational costs and enhances processing efficiency.
  3. Multi-Task Adaptability: InternW0 supports a variety of robotic operation tasks, including fine manipulation and scientific experiment workflows.

Applications

  • Scientific Experiment Automation: In complex scientific tasks such as metal-organic framework synthesis, InternW0 demonstrates its precise control capabilities across multiple stages.
  • Robotic Manipulation: In precision tasks like quantitative pipetting, InternW0 achieves high accuracy through contact sensing and force feedback mechanisms.

Industry Impact

The release of InternW0 marks a significant advancement in physical world models for real-world interactions. Its efficient processing capabilities and multi-task adaptability make it a valuable tool in robotics, autonomous driving, and smart manufacturing. Developers can utilize InternW0 to build smarter and more efficient robotic systems, driving further developments in human-machine interaction technologies.

Developer Recommendations

  • Understand the Model Architecture: Gain a deep understanding of InternW0's asymmetric video-action architecture and layer-wise K/V cache mechanism to fully leverage its efficient processing capabilities.
  • Integrate Multimodal Data: Combine multimodal data sources to enhance the model's adaptability and prediction accuracy in complex environments.
  • Continuous Optimization: Fine-tune the model according to specific application scenarios to achieve optimal performance.

Source: Hugging Face Daily Papers (2026-09-23)

— END —

Tags: #Shanghai AI Lab #InternW0 #Physical World Model #Robotic Control #Multimodal AI

Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.

What's your take on this story?
🚀 Explore & Submit Works

Community Comments

Loading live comments and annotations…