Table of Contents
Remember the 2010 Ranjnikanth starrer film Enthiran/Robot? Google Gemini’s AI is on the way to achieving what Dr. Vaseegaran had achieved with the fictional humanoid, Chitti! Google’s DeepMind division released a new large language model called Gemini Robotics On-Device that runs locally on robotic devices. In a blog post posted on 24 June 2025, Google says that the new AI model has been optimised to run effectively on humanoid robots, showcasing abilities like “strong general-purpose dexterity and task generalization”.
About Google Gemini AI
Gemini is Google’s latest and most powerful AI model, a successor to the previous model, Bard. It is defined as multimodal, meaning that it can understand and generate texts, images, videos, and audio. Gemini is considered to be the most powerful submission of Google in the AI race. Therefore, Gemini is built to compete with the likes of OpenAI’s ChatGPT and Perplexity AI.
The Gemini model is available in different sizes, from cloud-scale to mobile-ready versions. In essence, Gemini is not just a chatbot, rather a futuristic model that can integrate natural language understanding, logic, and reasoning across multiple platforms. In other words, a general-purpose intelligence layer that can interpret context, reason like a human, and now, power physical machines with cognitive skills.
The New Offline AI Model
The new offline AI model is built on the Gemini Robotics model, which was released earlier this year in March, and is named as Gemini Robotics On-Device model. This model can control a robot’s movements and can understand natural language prompts. The major highlight of this model is that it can perform without an internet connection. Google claims that this model is really useful in areas where there is zero connectivity and in applications that are latency sensitive.
The model is designed for robots with two arms (more like humans) and is engineered in a way that it requires “minimal computational resources” and can complete highly complicated or skilled tasks such as folding clothes, unzipping a bag, or putting random objects in a bag. You can watch this video to see the AI model in action.
Source: DeepMind
When it comes to completing complicated multi-step instructions and operating on out-of-distribution tasks, Google claims that Gemeni Robotics On-Device model exceeds and even outperforms other On-Device models. Though the model was trained to work on Aloha robots, the tech giant stated that its engineers adapted the new model and successfully worked it on a bi-arm Franka FR3 robot and an Apollo humanoid as well.
Additionally, on the bi-arm Franka FR3 robot, the model was able to follow general instructions and handle previously unseen objects, like executing an industrial belt assembly. Google also offers developers an opportunity to try out the Gemini Robotics On-Device using the software development kit (SDK).
Real-World Use Cases of Gemini-Powered Robots
So what exactly can robots that have Gemeni AI as brains do that makes this update and model interesting, or rather, how can it make our lives easier? The possibilities are endless and span several sectors.
Personal Assistants and Humanoids
Think of a physical Jarvis from the Ironman and Avengers franchise, where you get to have a humanoid who is your personal assistant. When not battling world-threatening aliens, humanoids powered by Gemini AI can execute your orders and interact with you and your family members without internet. This also means that the system does not store your conversations and routines in the cloud, thus emphasising privacy.
Healthcare Robots
In hospitals and clinics, robots could take up different roles and execute simple orders like navigating hallways, assist in patient care, record medical notes, and operate in data-sensitive areas without violating any privacy due to the lack of internet dependency.
Industrial Automation
Robots powered by on-device can sort items, navigate factory floors, coordinate with workers, and finally operate even when the network is down.
Rural or Field-based Robotics
Rural fields generally tend to have weaker signals of connectivity, resulting in the maximum utilisation of Gemeni AI-powered drones and robots. Besides inspecting the soil and planting seeds, they can also report data and adjust actions in real-time.
Conclusion: The Coming of Offline Robots
With each day passing by, our society is getting closer and closer to an existence where it is possible for humans to coexist with robots. Google isn’t the only tech giant that is working on AI models for robots. NVIDIA unveiled GrootN1, an AI model for humanoid robots, while HuggingFace is developing their robot powered by an in-house developed open-source model.
The AI race has led to many companies upping their AI game in innovative and relevant ways. With Google’s Gemini AI becoming a part of a physical robot, the idea and practicality of plugged robots or AI enclosed within lab walls are becoming a thing of the past.
Frequently Asked Questions
What is Gemini AI?
Gemini is Google’s next-gen multimodal AI model designed to power text, image, video, and audio-based tasks, now also used in robotics.
How is Gemini being used in robots?
Google is integrating Gemini into humanoid and mobile robots, allowing them to operate without internet by processing data directly on the device.
Are Gemini-powered robots available to consumers?
Not yet. Most implementations are in research and testing phases, but the technology is paving the way for consumer-level humanoids.
What makes offline AI important?
Offline AI ensures faster processing, improved privacy, and better functionality in areas without stable internet.
Is Gemini better than GPT-4?
It depends on the task. Gemini is competitive in multimodal understanding and on-device performance, while GPT-4 is widely used in language-specific applications.