DeepMind Announces AI Models to Enhance Physical Robots
Google DeepMind’s New AI Models for Robotics
Google DeepMind recently introduced two innovative artificial intelligence models designed specifically for robotics. These models, named Gemini Robotics and Gemini Robotics-ER (Extended Reasoning), are powered by Gemini 2.0, which Google refers to as its most advanced AI technology to date.
Partnership with Apptronik
To further develop these technologies, Google announced a partnership with Apptronik, a robotics company based in Texas. Apptronik is known for its past collaborations with significant organizations, including NASA and Nvidia. Recently, Google also participated in a $350 million funding round for Apptronik, emphasizing its commitment to supporting the next generation of humanoid robots that will leverage the capabilities of Gemini 2.0.
Demonstrations of Gemini Robotics
In demonstration videos released by Google, the Apptronik robots showcased their abilities using the new AI models. The robots were seen performing everyday tasks like plugging items into power strips, filling lunch boxes, moving plastic fruits and vegetables, and zipping bags—all executed in response to verbal commands. Despite the impressive capabilities demonstrated, Google has not yet provided a timeline for when these technologies will be commercially available.
Key Qualities of AI Models for Robotics
According to Google, there are three essential qualities that AI models must possess to be useful for robotics:
General Adaptability: These models need to adapt to various situations rather than being limited to specific tasks.
Interactivity: They should be capable of quickly understanding and responding to instructions and changes in their environment.
- Dexterity: This involves mimicking human-like skills, allowing robots to manipulate and interact with objects in a way that resembles human actions.
Gemini Robotics-ER for Developers
Gemini Robotics-ER is intended for developers in the robotics field, serving as a foundational tool for custom AI model training. This model is available not only to Apptronik but also to select testers, including companies like Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools.
The Competitive Landscape
Google is not alone in its endeavor to integrate AI into the field of robotics. In a significant move last November, OpenAI invested in a startup called Physical Intelligence, which aims to implement general-purpose AI into the physical realm by developing large-scale AI models for robotics applications. Furthermore, OpenAI has actively recruited experts, such as the former head of Meta’s augmented reality initiative, to spearhead its robotics projects.
Tesla is also making strides in humanoid robotics with its Optimus robot, indicating a rapidly evolving industry filled with significant players and innovations.
Comments from Google CEO
Sundar Pichai, the CEO of Google, expressed the company’s belief that robotics serves as an excellent testing platform for translating AI advancements into real-world applications. He highlighted the robots’ ability to leverage Google’s multimodal AI models, enabling them to adapt to their surroundings and make adjustments in real-time.
With this groundbreaking introduction of Gemini Robotics and the strategic partnership with Apptronik, Google aims to carve out a substantial presence in the robotics sector, setting the stage for a future where AI and robotics work hand in hand to assist in our daily lives.
