AI Robotics: Google DeepMind’s On-Device Model

As manufacturers worldwide pursue innovative strategies to integrate robotics and automation into their workflows, newly developed AI technology offers significant opportunities for improvement in efficiency, safety and economics.

Through the application of AI, robots can be empowered to perform tasks with advanced autonomous capabilities, boosting the operational potential for various industries.

An exciting development in this realm is Google DeepMind's introduction of Gemini Robotics On-Device, a language model (LM) that facilitates local operation on robots without requiring an internet connection.

A step forward for autonomous robotics

This on-device AI model is an advancement in autonomous robot control and adaptability.

It operates independently of internet connectivity, leveraging capabilities shaped by Google DeepMind's ongoing innovations.

Initiating with Gemini Robotics in March 2025, the purpose was to solve complex problems through multimodal reasoning across texts, images, audio and videos.

The evolution now integrates vision-language-action (VLA) capabilities—permitting enhanced autonomy.

The current iteration maintains dexterous functionalities akin to its predecessors.

Google emphasises its efficiency and minimal size, rendering it optimal for direct deployment on robots.

With Gemini Robotics On-Device, a robot's actions can be managed using natural language prompts, offering developers the flexibility to tweak and adapt the model for varied operational needs.

Benchmarking on-device performability

Google states that the on-device model presents performances close to the cloud-based Gemini Robotics model during tests.

Furthermore, they claim it surpasses other local models on general benchmarks, although specific competitor comparisons remain undisclosed.

Carolina Parada, Head of Robotics at Google DeepMind, acknowledges the model's unexpected strength: “The Gemini Robotics hybrid model is still more powerful, but we’re actually quite surprised at how strong this on-device model is.

“I would think about it as a starter model or as a model for applications that just have poor connectivity.”

Gemini Robotics supports a wide array of physical tasks without prior training.

It empowers robots to adapt dynamically, follow nuanced commands and execute tasks requiring intricate motor skills.

Demonstrations highlighted the model’s capabilities with tasks such as:

• Folding clothes
• Unzipping bags
• Pouring liquids
• Tying shoelaces

SDK launch for developer engagement

Alongside this model, Google is unveiling a software development kit (SDK).

This tool allows developers to refine and assess the on-device model’s applications on a physical simulator, reinforcing it as a first in Google DeepMind's VLA initiatives.

Google asserts that developers need only provide 50 to 100 task demonstrations to train robots utilising MuJoCo physics simulator models.

These offerings will initially be available to selected testers as Google strives to mitigate safety complications.

AI robotics: Broader applications and future prospects

The introduction of Gemini Robotics On-Device signifies an impactful step for AI-driven robotics.

As innovations persist, this technology promises extensive implications across various sectors globally:

• Remote Operations: Offline capabilities enhance robotic usage in limited connectivity scenarios—including space, disaster response, or regions with unreliable internet.
• Manufacturing and Logistics: Rapid adaptation to differing environments positions this model to transform warehouse and production line procedures.
• Healthcare: Local data processing secures privacy, enhancing use in sensitive settings like hospitals.

The continuous evolution of AI in robotics presents unparalleled opportunities for industries aiming to optimise their operational capacity and efficiency.

Explore the latest edition of AI Magazine and be part of the conversation at our global conference series, Tech & AI LIVE. Discover all our upcoming events and secure your tickets today.

Also sign up to our free weekly newsletter for the latest insights and stories straight into your inbox.

AI Magazine is a BizClik brand