Google DeepMind has introduced Gemini Robotics, a series of AI models designed to enhance robotic interaction with the physical world. Demonstration videos showcase robots executing tasks like folding paper and organizing objects based on voice commands. Trained to generalize behaviors across various hardware platforms, these models link visual recognition with actionable tasks. Notably, tests indicate that Gemini Robotics can perform effectively in unfamiliar environments. DeepMind has also released Gemini Robotics-ER for researchers to develop custom models and introduced the Asimov benchmark to assess AI robotics risks.
Google DeepMind's new AI models, Gemini Robotics, enable machines to interact with objects and navigate environments effectively, shown in various demo tasks.
Trained to generalize behavior across numerous hardware platforms, Gemini Robotics can link what robots 'see' to possible actions in real-world contexts.
In tests, Gemini Robotics performed well even in environments absent from its training, showcasing significant advancements in robotic interaction capabilities.
DeepMind provides a streamlined version, Gemini Robotics-ER, to help researchers develop their robotics control models and assess risks using the benchmark Asimov.
Collection
[
|
...
]