Robots are about to get a lot smarter—and they won’t need the internet to do it. Google DeepMind has announced Gemini Robotics On-Device, a version of its powerful Gemini AI that can run entirely on robots without relying on cloud connectivity.

This advancement means robots will be able to complete complex tasks, understand human commands, and adapt to new situations even in areas with poor or no internet connection. Google says the technology is designed for fast, reliable performance, making it ideal for real-world environments where lag or downtime could mean the difference between success and failure.

“Small and efficient enough to operate directly onboard,” is how Carolina Parada, head of robotics at Google DeepMind, describes the system. Unlike previous models that required a constant data connection to process complex instructions or learn new tasks, Gemini Robotics On-Device brings advanced reasoning and fine motor control to robots without constant oversight.

The system can be adapted with just 50 to 100 demonstrations, allowing developers to customize robots quickly for specific tasks without requiring vast amounts of training data. Initially trained on Google’s ALOHA robot, it has already been adapted to work on other platforms, including the Apptronik Apollo humanoid and the Franka FR3.

For the first time, Google is allowing developers to fine-tune a DeepMind robotics model for their needs. The company has released a full SDK to support experimentation and has opened a trusted tester program to allow developers early access to test the system in live environments.

Running AI directly on robots offers significant privacy advantages. All data remains local, which is crucial for sensitive tasks in healthcare, security, and personal assistance. It also means robots can continue functioning during internet outages or in remote areas, unlocking new opportunities for using robots in disaster zones, rural regions, and infrastructure-poor environments.

Google’s push toward local processing also helps reduce latency, delivering faster response times and fewer points of failure. This makes robots more reliable in critical tasks where every second counts.

However, Google has flagged that the on-device system does not come with built-in semantic safety features. The company is urging developers to integrate safety protocols into their robots using tools like the Gemini Live API and trusted low-level controllers. For now, the technology is being limited to select developers to carefully monitor safety risks before wider release.

While hybrid models that use a mix of cloud and on-device AI still offer more raw power, the on-device model is designed to handle most common robotic use cases without needing constant updates or data streams from a server farm.

In a world increasingly driven by AI, Google’s Gemini Robotics On-Device could mark a turning point in how robotics are integrated into daily life. By cutting the dependency on cloud systems, robots gain more autonomy and can operate in a broader range of environments.

The technology’s arrival could have significant implications for industries ranging from manufacturing and logistics to healthcare and elder care. Imagine robots assisting in hospitals without risking data leaks, or aiding in warehouse operations even if the network goes down, or helping rescue teams in disaster-stricken areas where connectivity is unreliable.

Google’s move reflects a growing industry trend toward edge computing, where devices process data locally instead of relying on centralized servers. By bringing Gemini’s advanced reasoning into an offline format, Google is positioning robots to become more adaptive, secure, and practical for real-world applications.

As robots equipped with this technology begin to roll out, it could reshape expectations around what robots can achieve—and where they can function effectively—pushing robotics closer to becoming an everyday presence in homes and workplaces.