DeepMind Announces Plans to Utilize AI Models for Physical Robots
Google DeepMind Launches New AI Models for Robotics
Google DeepMind recently introduced two innovative AI models designed for use in robotics, running on their latest framework called Gemini 2.0. This represents what Google has termed its most advanced AI technology to date, geared towards enhancing robotic functionalities.
New AI Models: Gemini Robotics and Gemini Robotics-ER
The two models debuted by Google are Gemini Robotics and Gemini Robotics-ER (Extended Reasoning). Unlike traditional generative AI that primarily produces text and images, Gemini Robotics focuses on commanding physical actions in robots, marking a significant step forward in the integration of AI into physical tasks.
Partnership with Apptronik
To advance its efforts in robotics, Google announced a partnership with Apptronik, a robotics company based in Texas. This collaboration aims to develop the next generation of humanoid robots equipped with Gemini 2.0 technology. Apptronik has previously worked with notable organizations like Nvidia and NASA, and they recently secured $350 million in funding, with Google participating in this investment round.
Demonstration of Capabilities
In demonstration videos shared by Google, Apptronik robots showcased a range of tasks guided by the new AI models. They were seen plugging devices into power strips, organizing lunchboxes, and manipulating plastic vegetables—all in response to voice commands. However, Google did not specify when these robotic technologies will be available for public use.
Key Characteristics of the AI Models
Google emphasizes three essential qualities for any AI models used in robotics:
- General Adaptability: The robots should adapt to various situations and tasks, showcasing flexibility.
- Interactivity: These models must understand and react promptly to verbal instructions and environmental changes.
- Dexterity: The robots need to exhibit capabilities akin to human hands, allowing them to handle objects carefully.
Gemini Robotics-ER: A Tool for Developers
Gemini Robotics-ER serves as a foundational platform for developers to train their custom models. This version of the AI is accessible not only to Apptronik but also to select trusted testers, including Agile Robots, Boston Dynamics, and Enchanted Tools, enabling a broader base for innovation and development.
Competitive Landscape in AI Robotics
Google is not the only player focusing on AI for robotics. In a similar realm, OpenAI has made investments in Physical Intelligence, a startup dedicated to integrating general-purpose AI within physical environments. This startup aims to develop substantial AI models and algorithms to empower robots.
Alongside OpenAI’s pursuits, Tesla is also entering the humanoid robotics field with its Optimus robot, further intensifying competition in the industry.
Insights from Google’s Leadership
Google CEO Sundar Pichai recently noted the company’s vision of robotics as an essential platform for applying advances in AI to real-world situations. He stated that the robots would utilize Google’s multimodal AI models to adapt and modify their actions in real time based on their surroundings.
With this recent launch, Google aims to push the boundaries of robotics further, setting the stage for a future where AI technologies seamlessly integrate into everyday physical tasks. This initiative not only reflects technological advancements but also highlights the increasing convergence of AI and robotics in various domains.
