InternVLA-N1 - Shanghai AI Lab Open Source End-to-End Dual System Navigation Large Model

What is InternVLA-N1?

InternVLA-N1 is an open source end-to-end dual-system navigation macromodel from Shanghai Artificial Intelligence Laboratory. It adopts a dual-system architecture, where System 2 is responsible for understanding linguistic commands and planning long-range paths, and System 1 focuses on high-frequency response and agile obstacle avoidance. The model is trained entirely based on synthetic data, and a cost-effective training process is realized through large-scale digital scene assets and massive multimodal corpus. In several mainstream benchmark tests, InternVLA-N1 performs well and scores at the international leading level, demonstrating strong zero-sample generalization capability. InternVLA-N1 is able to realize "long-distance walking across buildings" and agile obstacle avoidance among dense obstacles in real-life scenarios.

Functional Features of InternVLA-N1

• Dual system co-navigation: Combine the long-range spatial reasoning of System 2 with the high-frequency obstacle avoidance capability of System 1 to achieve efficient and stable navigation.
• Pure synthetic data training: Relying only on synthetic data to complete training, dramatically reducing data acquisition costs while maintaining excellent performance.
• Zero-sample generalization capability: Quickly adapt and accurately perform navigation tasks in new environments without additional real-world scenario data.
• High Frequency Response and Agile Obstacle Avoidance: System 1 responds to environmental changes at 60Hz, ensuring flexible obstacle avoidance in complex scenes.
• multimodal interaction: Supports verbal command input and visual environment perception, realizing natural and smooth human-computer interaction.
• Cross-scenario generalization: Wide range of applicability for a variety of complex scenarios, such as indoor, outdoor, and between buildings.
• Open Source and Scalability: Provide open source code and models to facilitate secondary development and application expansion by developers.

Core Benefits of InternVLA-N1

• Efficient navigation capabilities: The organic combination of long-range planning and high-frequency obstacle avoidance is realized through the cooperative work of the dual systems, which significantly improves the navigation efficiency and stability.
• Data-driven low-cost training: Training is based entirely on synthetic data, eliminating the need for large amounts of real data collection and dramatically reducing training costs and time.
• Strong generalization capabilities: performs well in real environments with strong zero-sample generalization even when trained using only synthetic data.
• High Frequency Response and Agile Obstacle Avoidance: System 1 responds to environmental changes at a frequency of 60Hz, ensuring flexible obstacle avoidance and adaptation to dynamic environments in complex scenes.
• Multimodal Interaction Support: Supports verbal command input and visual environment perception, realizing natural and smooth human-computer interaction and enhancing user experience.
• Wide range of applicability: It is highly versatile and suitable for a wide range of complex scenarios, such as indoor, outdoor, and between buildings.

What is InternVLA-N1's official website?

• Project website:: https://internrobotics.github.io/internvla-n1.github.io/
• GitHub repository:: https://github.com/InternRobotics/InternNav
• Hugging Face Model Library:: https://huggingface.co/InternRobotics/InternVLA-N1
• Technical Papers:: https://internrobotics.github.io/internvla-n1.github.io/static/pdfs/InternVLA_N1.pdf

Applicable People of InternVLA-N1

• Artificial intelligence researchers: Researchers interested in large models of navigation who wish to study and improve navigation algorithms.
• Robotics Engineer: Engineers who develop service robots, logistics robots, etc. that require navigation functions.
• Autonomous Driving Developers: Developers engaged in the development of autonomous driving technologies that require efficient path planning and obstacle avoidance functions.
• Virtual Reality/Augmented Reality Developer: Developers who need to implement natural interaction and navigation features in VR/AR applications.
• Smart Security Practitioners: Practitioners in the field of smart security who need to realize smart patrols and abnormal responses.
• Industrial Automation Engineer: Engineers who need to automate equipment navigation and operation in industrial environments.