DeepMind Announces Plans to Utilize AI Models for Physical Robots
Google DeepMind and Its New AI Models for Robotics
Introduction to Gemini 2.0
On Wednesday, Google DeepMind introduced two innovative AI models specifically designed for robotics. Named Gemini Robotics and Gemini Robotics-ER (Extended Reasoning), these models are powered by Gemini 2.0, which Google describes as its most capable AI to date. This technological advancement represents a significant shift, taking AI beyond just text and visuals into the realm of physical actions that can be executed by robots.
Collaboration with Apptronik
Google has announced its partnership with Apptronik, a robotics company based in Texas. This collaboration aims to develop the next generation of humanoid robots using Gemini 2.0. Apptronik has a rich portfolio, having previously worked with notable organizations such as Nvidia and NASA. Recently, they also secured substantial funding, with Google participating in a $350 million funding round that will enhance their robotics initiatives.
Demonstrations of New Capabilities
In several demonstration videos, Google showcased the capabilities of Apptronik robots powered by the new AI models. These robots successfully performed a variety of tasks, including:
- Connecting devices to power strips
- Packing a lunchbox
- Arranging plastic vegetables
- Zipping up bags
The robots responded to verbal commands, demonstrating how Gemini Robotics models can interpret and act on instructions in real-time. However, Google has not yet announced when these advanced robotic technologies will be available for public use.
Essential Qualities of AI in Robotics
For AI models in robotics to be effective, Google emphasizes the importance of three key qualities:
- General Adaptability: Robots should be able to adjust to different situations and environments.
- Interactive Response: They must quickly understand and respond to new instructions or changes around them.
- Dexterity: Robots should have the ability to manipulate objects in a manner similar to human hands.
These qualities are essential for the robots to be genuinely useful and helpful in everyday scenarios.
Features of Gemini Robotics-ER
The Gemini Robotics-ER model is specifically crafted for roboticists, providing a foundation for them to train their own models. Google has made this model available to Apptronik and a select group of trusted testers, which includes prominent robotic companies like Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools. This exclusive access aims to refine and enhance the capabilities of the Gemini models.
Competing Innovations in Robotics
Google is not the only tech giant exploring the integration of AI into robotics. In November, OpenAI made headlines by investing in Physical Intelligence, a startup focused on developing general-purpose AI for robotics. This startup aims to provide large-scale AI models to power robots, expanding the capabilities of AI in physical applications.
Additionally, OpenAI has made strategic hires for its robotics initiatives, including the former head of Meta’s augmented reality project. Tesla is also venturing into humanoid robotics with its Optimus robot, which showcases the growing interest in this dynamic field.
Vision and Adaptability in Robotics
During the announcement, Google’s CEO Sundar Pichai expressed a vision of robotics acting as a testing ground for applying AI advancements in real-world settings. He noted that these robots, powered by Google’s multimodal AI models, would be able to adapt and make changes on-the-fly, showcasing a level of flexibility that is crucial for future applications.
As the sector evolves, these partnerships and technological advancements from companies like Google, OpenAI, and Tesla signify a robust competition in developing effective and versatile robotic systems. The ongoing innovations reflect a promising future for AI-assisted robotics, with potential applications in diverse industries from healthcare to manufacturing.
