DeepMind Announces Plans to Utilize AI Models for Enhancing Physical Robots
Google DeepMind Unveils New AI Models for Robotics
Introduction to Gemini Models
On Wednesday, Google DeepMind revealed two new artificial intelligence models specifically designed for robotics, which are powered by Gemini 2.0. Google refers to this as its "most capable" AI model to date. These models signify an important step forward, expanding the range of generative AI that has previously been limited to producing text and images into the realm of physical robot commands.
Partnership with Apptronik
In a significant move, Google announced a collaboration with Apptronik, a robotics developer based in Texas. The aim is to create the next generation of humanoid robots using the capabilities of Gemini 2.0. Historically, Apptronik has collaborated with high-profile entities such as Nvidia and NASA, further underscoring the innovative potential of this partnership. Recently, Google participated in a $350 million funding round for Apptronik, highlighting its commitment to advancing humanoid robotics.
Demonstration of Robotic Capabilities
In a series of demonstration videos, Google showcased Apptronik’s robots utilizing the Gemini AI models. These robots performed a variety of tasks, including:
- Plugging devices into power strips
- Filling lunchboxes with items
- Handling plastic vegetables
- Zipping up bags
These actions were executed in response to spoken commands, showcasing the robots’ responsiveness and adaptability.
Essential Qualities of AI Models for Robotics
According to Google, effective AI models for robotics must possess three key qualities:
- General Adaptability: They should be capable of adjusting to various situations as required.
- Interactive Capability: They need to understand and promptly respond to user instructions and changes in their environment.
- Dexterity: They must be able to perform tasks that typically require human-like manipulation, such as handling objects carefully.
Gemini Robotics-ER — A Tool for Developers
In addition to the main Gemini Robotics model, Google has introduced Gemini Robotics-ER, tailored for roboticists to develop their AI models further. This platform is made available not only to Apptronik but also to "trusted testers" in the robotics field, including companies like Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools. This collaborative approach is likely to foster innovation and create more versatile robotic solutions.
The Broader AI Robotic Landscape
Google is part of a larger trend where tech companies are exploring the intersection of AI and robotics. In November, OpenAI made an investment in a startup named Physical Intelligence. This company aims to introduce general-purpose AI into the physical realm with the advancement of large-scale AI models and algorithms for robots.
Additionally, OpenAI recently hired a former executive from Meta’s Orion project to oversee its robotics initiatives, which reflects the growing interest among tech giants in this space. Tesla is also playing a role in this evolution by developing its humanoid robot, known as Optimus.
Google’s Vision for Robotics
Sundar Pichai, the CEO of Google, shared his vision for the integration of AI into the physical world via robotics. In a post on social media platform X, he highlighted the importance of robotics as a testing ground for AI developments. He emphasized that Google’s AI models would allow robots to adapt in real-time and respond effectively to their surroundings, making them more intuitive and functional.
This initiative marks a significant step toward merging advanced AI capabilities with practical applications in robotics, paving the way for future developments in this exciting field.
