DeepMind Announces AI Models for Enhancing Physical Robots
Google DeepMind Introduces New AI Models for Robotics
Google DeepMind has recently unveiled two innovative AI models designed specifically for robotics, which operate on the company’s latest release, Gemini 2.0. Dubbed "the most capable" AI model to date by Google, these advancements signal a significant leap in the integration of artificial intelligence into physical tasks performed by robots.
Key Features of Gemini Robotics
AI Models:
- Gemini Robotics: This model moves beyond the typical functions of generative AI, which predominantly handles text and imagery. It is tailored to translate digital commands into real-world actions for robotic systems.
- Gemini Robotics-ER (Extended Reasoning): A specialized model aimed at providing roboticists with a foundational tool for developing their own AI systems.
- Demonstrations:
In several demonstration videos, robots from Apptronik, a Texas-based robotics company, showcased the capabilities of these new models. Actions included plugging devices into power strips, organizing lunchboxes, and maneuvering plastic vegetables—all accomplished in response to spoken instructions.
Google’s Collaboration with Apptronik
Google has announced a partnership with Apptronik to create advanced humanoid robots utilizing the capabilities of Gemini 2.0. Apptronik has established itself in the robotics field, having previously collaborated with notable organizations like Nvidia and NASA. In a recent funding round, Google played a pivotal role by contributing to Apptronik’s $350 million investment effort.
Essential Qualities of Robotics AI Models
According to Google, successful AI models for robotics should fulfill three critical criteria:
- Adaptability: These models must be able to navigate and adjust to various unexpected scenarios.
- Interactivity: The ability to quickly comprehend and respond to verbal commands and environmental changes is essential.
- Dexterity: The robots should perform tasks typically done by human hands, such as manipulating small objects with care.
Gemini Robotics-ER for Developers
The Gemini Robotics-ER model is specifically designed for robotic developers to train and develop customized applications. It is made accessible to various trusted partners, including Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools. This collaboration aims to foster innovation in the field by leveraging advanced AI capabilities.
The Competitive Landscape
Google is not alone in its pursuit of integrating AI into robotics. Recently, OpenAI invested in a startup called Physical Intelligence, which focuses on translating general-purpose AI into the robotic domain. The investment aligns with OpenAI’s vision of developing algorithms that enhance robot performance in physical tasks.
Tesla is also making strides within the humanoid robotics sector. Their humanoid robot, named Optimus, aims to revolutionize the production of robots designed for intricate functionalities.
Insights from Google CEO
In remarks shared on social media, Google CEO Sundar Pichai emphasized that robotics serves as an essential testing ground for advancing AI applications in real-world contexts. He pointed out that these robots will utilize Google’s multimodal AI models to adapt and dynamically respond to their environments, enhancing their usability in everyday tasks.
The future of robotics looks promising as AI capabilities continue to evolve, paving the way for smarter systems that can seamlessly interact with their surroundings and assist humans in various tasks. This development reflects the continuous efforts of major tech companies to innovate and expand the possibilities of artificial intelligence in our daily lives.
