Google Deepmind is launching two new Artificial Intelligence (AI) models, Gemini Robotics and Gemini Robotics-ER. These latest iterations are powered by Google’s Gemini 2.0, setting the foundation for the next generation of assistive robots.
The innovation is set to solve the gap in the abilities of the company’s existing robots, which are limited to the digital realm. Consequently, the new versions bring the power of AI into real-life, physical applications.
According to Google’s AI research lab, Gemini Robotics and Gemini Robotics-ER possess the necessary characteristic to advance what their earlier models can do. Specifically, they have embodied reasoning to interact like a real human with the world around us and be able to do things with safety.
Gemini Robotics
Deepmind describes Gemini Robotics as an advanced vision-language-action (VLA) model. It also integrated a new output modality to control the robots directly and conduct physical actions.
The first new AI model represents notable improvements in three key aspects, including generality, interactivity, and dexterity. Gemini Robotics can smoothly adapt to various situations and robot types. They can easily follow instructions and respond to environmental changes. Moreover, the model is capable of doing things human-like, specifically object manipulation.
Gemini Robotics-ER
Google Deepmind says that Gemini Robotics-ER allows roboticists to execute and manage their own programs.
Focusing on spatial reasoning, Gemini Robotics-ER heightens Gemini’s understanding of the world. The latest version significantly improves pointing and 3D detection features. It can also instantly generate fresh capabilities when encountering new scenarios.
Both Gemini Robotics and Gemini Robotics-ER push robots to perform better in various real-world activities.
In addition, the research team is planning to further explore the capabilities of the new models to continue the development.
“As we explore the continuing potential of AI and robotics, we’re taking a layered, holistic approach to addressing safety in our research, from low-level motor control to high-level semantic understanding,” Carolina Parada added.
Meanwhile, Google Deepmind is also partnering with Apptronik to produce new generations of humanoid robots.Â
