Google’s AI division, DeepMind, has introduced new models for robot control based on Gemini 2.0.
These models enable machines to interact with real-world objects, navigate their surroundings, and perform various tasks.
Gemini Robotics is an advanced neural network that incorporates physical actions as output information for robot control. Gemini Robotics-ER is a model with enhanced spatial understanding.
Both systems allow robots to perform a wide range of real-world tasks. The team has released a series of demonstration videos showing them folding paper, placing glasses in a case, and completing other tasks in response to voice commands.
DeepMind noted that during tests, the robots operated in conditions not included in the training data. Developers have released a scaled-down version of Gemini Robotics-ER, which other researchers can use to train their own robot control models.
In March, the company Agibot launched the AI Genie Operator-1 (GO-1) for humanoid robots, enabling rapid comprehension and task execution.
Earlier, Figure introduced its own artificial intelligence, Helix, for integration with mechanical devices. According to its creators, the model is capable of “reasoning like a human.”