DeepMind Plans to Utilize AI Models for Enhancing Physical Robots
Google Unveils Advanced AI Models for Robotics
Introduction to Gemini 2.0
On Wednesday, Google DeepMind announced the launch of two innovative AI models designed specifically for robotics. Both models operate on Gemini 2.0, which Google describes as its "most capable" AI technology yet. These models are intended to enhance the capabilities of robots, enabling them to perform physical tasks through precise action commands, rather than just generating text or images.
Partnership with Apptronik
Google has teamed up with Apptronik, a Texas-based robotics developer, to create the next generation of humanoid robots that will utilize Gemini 2.0. This collaboration will help in the development of robots that can engage in a multitude of tasks around the home or workplace. Apptronik has an impressive background, having collaborated with industry giants like Nvidia and NASA.
Recently, Google participated in Apptronik’s $350 million funding round to support their innovative work in robotics. The company aims to build robots that not only replicate human actions but also respond intelligently to verbal instructions.
Capabilities of Gemini Robotics
In a series of demonstration videos, Google showcased Apptronik robots equipped with the new AI models. These performances highlighted the robots’ ability to perform tasks such as:
- Plugging items into power outlets
- Filling lunchboxes
- Moving objects like plastic vegetables
- Zipping up bags
These demonstrations illustrated how the robots respond to spoken commands, showcasing their potential usability in everyday life. However, Google has yet to specify when this technology will be available to the public.
Essential Qualities for Robotics AI
For AI to be effective in robotics, Google outlined three critical qualities that these models must possess:
General Adaptability: Robots should be able to handle various situations, adjusting as needed.
Interactivity: They need to understand and respond quickly to instructions or alterations in their environment.
- Dexterity: The ability to perform tasks similar to those humans do with their hands, like careful manipulation of objects.
These attributes are crucial for making AI models truly useful in real-world applications.
Introducing Gemini Robotics-ER
Another notable model introduced is Gemini Robotics-ER, tailored specifically for roboticists aiming to develop their own models. This tool is available not only to Apptronik but also to select "trusted testers," including Agile Robots, Boston Dynamics, and Enchanted Tools. This approach fosters collaboration within the robotics community to innovate and improve AI applications in robotics.
The Competitive Landscape in AI Robotics
Google is not the only entity exploring AI for robotics. In November, OpenAI invested in Physical Intelligence, a startup focused on integrating general-purpose AI into robots. The startup is developing AI algorithms and models that can be applied to various robotic tasks.
Additionally, OpenAI recently hired a lead from Meta’s augmented reality initiative to spearhead its robotics and consumer hardware efforts. Similarly, Tesla is actively pursuing advancements in humanoid robotics with its own project known as the Optimus robot.
Future of Robotics at Google
Google’s CEO, Sundar Pichai, emphasized the company’s vision for robotics during the announcement. He mentioned that Google views robotics as an important area for testing and applying the latest advancements in AI to the physical world. The platform’s capability to adapt and make real-time changes in response to surroundings is expected to enhance the functionality of future robots.
By concentrating on making AI impactful in tangible settings, Google aims to pave the way for intelligent robots that can assist in everyday tasks, opening doors to new opportunities in various fields, including healthcare, customer service, and home assistance.
