Gemini Robotics On-Device

Gemini Robotics On-Device is Google's powerful VLA model, optimized to run locally on robots. It offers strong dexterity and fast task adaptation with minimal fine-tuning.

**Gemini Robotics On-Device** is Google's advanced AI model designed to run locally on robotic devices, offering high-performance dexterity and quick task adaptation without relying on cloud connectivity. Optimized for low-latency inference, it excels in complex, real-world tasks like object manipulation and multi-step instructions, even in environments with poor or no internet access. The model supports rapid fine-tuning with minimal demonstrations, making it adaptable to diverse robotic systems, including bi-arm and humanoid robots. Developers can leverage the included SDK to test and customize the model for specific applications using a physics simulator. Built with safety and responsibility in mind, Gemini Robotics On-Device aligns with Google’s AI Principles, ensuring robust performance while minimizing risks. Currently available through a trusted tester program, it aims to accelerate robotics innovation by addressing latency and connectivity challenges.