Gemini Robotics On-Device brings AI to local robotic devices

Details

Gemini Robotics On-Device is DeepMind's VLA (Vision-Language-Action) model optimized for local execution on robotic hardware without requiring cloud connectivity during operation. The model demonstrates strong dexterity and task generalization, following natural language instructions for complex manipulation tasks across diverse scenarios. Running on-device addresses a critical gap in practical robotics deployment: network-dependent models fail in environments with limited or unreliable connectivity — warehouses, construction sites, remote field operations. The Gemini Robotics SDK provides developers with tools for task evaluation, fine-tuning, and deployment integration. DeepMind positions this as commercially deployable rather than a research model, making it a meaningful step toward practical robotics applications using large vision-language models at the edge.