New Google AI makes robots smarter without the cloud

Google DeepMind has introduced a powerful on-device version of its Gemini Robotics AI. This new system allows robots to complete complex tasks without relying on a cloud connection. Known as Gemini Robotics On-Device, the model brings Gemini’s advanced reasoning and control capabilities directly into physical robots. It is designed for fast, reliable performance in places with poor or no internet connectivity, making it ideal for real-world, latency-sensitive environments.

 

 

 

Smarter robots that work anywhere

Unlike its cloud-connected predecessor, this version runs entirely on the robot itself. It can understand natural language, perform fine motor tasks, and generalize from very little data, all without requiring an internet connection. According to Carolina Parada, head of robotics at Google DeepMind, the system is “small and efficient enough” to operate directly onboard. Developers can use the model in situations where connectivity is limited, without sacrificing intelligence or flexibility.

 

QUADRUPED ROBOT PLAYS BADMINTON WITH YOU USING AI

 

Easy to adapt and train

Gemini Robotics On-Device can be customized with just 50 to 100 demonstrations. The model was first trained using Google’s ALOHA robot, but it has already been adapted to other platforms like Apptronik’s Apollo humanoid and the Franka FR3. For the first time, developers can fine-tune a DeepMind robotics model. Google is offering access through its trusted tester program and has released a full SDK to support experimentation and development.

 

HUMANOID ROBOTS HANDLE QUALITY CHECKS AND ASSEMBLY AT AUTO PLANT

 

Local control means more privacy and reliability

Since the AI runs directly on the robot, all data stays local. This approach offers better privacy for sensitive applications, such as in healthcare. It also allows robots to continue operating during internet outages or in isolated environments. Google sees this version as a strong fit for remote, security-sensitive, or infrastructure-poor settings. The system delivers faster response times and fewer points of failure, opening up new possibilities for robot deployment in real-world settings.

 

Safety requires developer input

The on-device model does not include built-in semantic safety features. Google recommends that developers build safety systems into their robots using tools like the Gemini Live API and trusted low-level controllers. The company is limiting access to select developers to better study safety risks and real-world applications. While the hybrid model still offers more overall power, this version holds its own for most common use cases and helps push robotics closer to everyday deployment.

 

AI TENNIS ROBOT COACH OFFERS PRO-LEVEL TRAINING ANYTIME, ANYWHERE

 

Kurt’s key takeaways

The release of Gemini Robotics On-Device marks a turning point. Robots no longer need a constant cloud connection to be smart, adaptive, and useful. With faster performance and stronger privacy, these systems are ready to tackle real-world tasks in places where traditional robots might fail.

Would you be comfortable handing off tasks to a robot that doesn’t need the internet to think? Let us know your thoughts in the comments below.