DeepMind Announces Plans to Utilize AI Models for Enhancing Physical Robots
Google DeepMind Unveils New AI Models for Robotics
Google DeepMind has recently introduced two innovative AI models aimed at revolutionizing the field of robotics. Both models, named Gemini Robotics and Gemini Robotics-ER (Extended Reasoning), operate on the highly advanced Gemini 2.0, which Google describes as its "most capable" AI to date. The focus of these models is not only on generating text and images but also on controlling robots through physical action commands.
Partnership with Apptronik
In a significant move to advance humanoid robotics, Google has entered into a collaboration with Apptronik, a robotics development company based in Texas. This partnership aims to create the next generation of humanoid robots powered by the Gemini 2.0 models. Apptronik is known for its previous collaborations with major organizations such as Nvidia and NASA. Recently, Google participated in Apptronik’s $350 million funding round, showcasing its commitment to the project.
Demonstration of Capabilities
During a series of demonstration videos, Google showcased the capabilities of Apptronik robots using the new AI models. Some of the impressive tasks performed by these robots included:
- Plugging items into power strips
- Filling lunchboxes
- Moving plastic vegetables
- Zipping up bags
These actions were executed in response to voice commands, highlighting the advanced interaction capabilities of the Gemini AI technology. Although many aspects of the models are demonstrated, Google has yet to provide a specific timeline for when these technologies will be available for commercial use.
Core Features of Gemini Robotics Models
To ensure that AI models for robotics are effective, Google highlighted three essential qualities they must possess:
- General Adaptation: The ability to adjust to different situations.
- Interactive Response: Quick understanding and responsiveness to instructions or environmental changes.
- Dexterity: The capability to manipulate objects similarly to how humans use their hands and fingers.
These attributes are vital for creating robots that can operate effectively in various environments and tasks.
Gemini Robotics-ER: A Model for Developers
The Gemini Robotics-ER model is designed with roboticists in mind. It serves as a foundational tool for them to develop and train their own robotics models. This model is available not only to Apptronik but also to a select group of “trusted testers,” which includes companies such as Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools.
Competition in the Robotics Arena
Google’s endeavor is not isolated, as the field of robotics with AI integration is rapidly expanding. For instance, OpenAI has recently invested in a startup called Physical Intelligence, which focuses on incorporating general-purpose AI into the physical world.
In conjunction with this development, OpenAI has brought on a prominent leader from Meta’s augmented reality division to spearhead its robotics initiatives. Additionally, Tesla is making strides in the humanoid robotics market with its own Optimus robot.
A New Era for Robotics
Sundar Pichai, CEO of Google, expressed that the company views robotics as a valuable testing ground for applying advancements in AI. According to Pichai, the robots equipped with Google’s multimodal AI models will be capable of making real-time adjustments and adapting to their surroundings effectively.
The ongoing progress in this field signifies a growing interest in how AI can transform robotics, opening up possibilities for more intelligent and responsive machines in our daily lives. As these technologies develop, they hold the potential to reshape various industries and enhance our interaction with machines.
