Google's new Gemini AI can power robots and make them work without internet

India Today

4 days ago

Google DeepMind has launched a new version of its Gemini Robotics AI model that allows robots to operate entirely without internet access. Called Gemini Robotics On-Device, the system is designed to power robots in real-world settings where speed, autonomy, and privacy are crucial. This update marks a significant shift from earlier models that relied on cloud connectivity. By enabling robots to process information and make decisions on the device itself, Google hopes to make robotics more practical in offline environments such as remote areas, secure facilities, and latency-sensitive situations.advertisement'It's small and efficient enough to run directly on a robot,' said Carolina Parada, head of robotics at Google DeepMind, in a statement to The Verge. 'I would think about it as a starter model or as a model for applications that just have poor connectivity.'Despite being a smaller variant, the on-device version holds its own. 'We're actually quite surprised at how strong this on-device model is,' Parada added.Gemini Robotics On-Device brings several new features to the table. The model can carry out tasks straight out of the box and learn new ones from as few as 50 to 100 demonstrations. The model was initially trained using Google's ALOHA robot, but it has since been successfully adapted for use with other robotic systems, such as Apptronik's Apollo humanoid and the dual-armed Franka FR3.advertisement
Google says that it can perform detailed actions such as folding clothes or unzipping bags, all while running low-latency inference on-device.For perspective, Tesla's humanoid robot, Optimus, can also do all those things – folding a t-shirt, boiling an egg, dancing, etc – but it needs an internet connection to send data to cloud servers.
However, in the case of Gemini Robotics On-Device, a standout feature is that all data is processed locally. That makes it particularly useful for privacy-sensitive applications, such as healthcare and industrial automation, where data security is a concern.'When we play with the robots, we see that they're surprisingly capable of understanding a new situation,' said Parada, highlighting the model's flexibility and responsiveness.Because the system does not rely on the cloud, it also keeps functioning in places with weak or no connectivity, making it highly reliable. 'It's drawing from Gemini's multimodal world understanding in order to do a completely new task,' Parada explained.However, unlike the cloud-based hybrid version, the on-device model does not include built-in semantic safety tools. Google recommends that developers implement their own safety systems, including using Gemini Live APIs and connecting to low-level safety controls.advertisement'With the full Gemini Robotics, you are connecting to a model that is reasoning about what is safe to do, period,' said Parada.This launch comes shortly after Google introduced the AI Edge Gallery, an Android app that lets users run AI models offline on their smartphones. Powered by the compact Gemma 3 1B model, the app allows users to generate images, write text, and interact with AI tools directly on their devices – no internet required.Just like Gemini Robotics On-Device, AI Edge Gallery focuses on privacy and low-latency performance. It uses open-source models from platforms like Hugging Face and technologies like TensorFlow Lite to ensure smooth experiences across a range of devices.- Ends