Google DeepMind officially introduces Gemini Robotics On-Device, optimized to run on efficiently on the robot itself
Google DeepMind yesterday introduced the Gemini Robotics On-Device, VLA model, optimized to run locally on robotic devices.
Today, we introduced Gemini Robotics On-Device 🤖
🧪 Designed for rapid experimentation with dexterous manipulation.
🦾 Adaptable to new tasks through fine-tuning to improve performance.
👟 Optimized to run locally with low-latency inference.Learn more… https://t.co/h2d1TZ49qm
— Google AI (@GoogleAI) June 24, 2025
Gemini Robotics On-Device
It is a robotics foundation model for bi-arm robots, engineered to require minimal computational resources. It builds on the task generalization and dexterity capabilities of Gemini Robotics and is:
- Designed for rapid experimentation with dexterous manipulation.
- Adaptable to new tasks through fine-tuning to improve performance.
- Optimized to run locally with low-latency inference.
It follows natural language instructions and completes highly dexterous tasks like unzipping bags or folding clothes, all while operating directly on the robot. The company has also revealed this new model’s generalization performance, compared to flagship Gemini Robotics models and the previous best on-device model via a chart.
The company is making the model available for fine-tuning. While many tasks will work out of the box, developers can also choose to adapt the model to achieve better performance for their applications. This model quickly adapts to new tasks, with as few as 50 to 100 demonstrations. Gemini Robotics On-Device model was adapted to different robot embodiments- ALOHA robots, Franka FR3 robots, and Apollo humanoid robots.
To gain a deeper understanding of this model’s usage and safety profile and to gather feedback, it is initially released to a select group of trusted testers. Meanwhile, a Gemini Robotics SDK is also being shared to help developers easily evaluate this model in terms of their tasks and environments. Developers can access the SDK by signing up for Google’s trusted tester program.
