# Google DeepMind Unveils Enhanced AI Model for Robotics - ForkLog

*Источник: ForkLog*
*Дата: 2026-07-31*
*Язык: en*

**Кратко:** Google DeepMind Unveils Enhanced AI Model for Robotics
Google DeepMind unveils Gemini Robotics 2 for full-body robot control. Google DeepMind has introduced Gemini Robotics 2, a Vision-Language-Action (VLA) model for full-body robot control, along with the ER 2 solution for comprehensive planning and coordination of multiple devices.

Google DeepMind Unveils Enhanced AI Model for Robotics
Google DeepMind unveils Gemini Robotics 2 for full-body robot control.
Google DeepMind has introduced Gemini Robotics 2, a Vision-Language-Action (VLA) model for full-body robot control, along with the ER 2 solution for comprehensive planning and coordination of multiple devices.
Gemini Robotics 2 translates text and visual data into motor actions. The model is designed for robots of various form factors, from desktop manipulators to full-sized humanoids.
Unlike the previous version, the AI now controls the entire body. This enables robots to walk, squat, bend, reach, and lift objects in confined spaces.
In demonstrations, the humanoid Apptronik Apollo 2, powered by Gemini Robotics 2, picked up a watering can, removed a baseball glove from a shelf, and located specific items.
Google DeepMind also announced improved motor skills: the model supports complex hands with 22 degrees of freedom and can perform tasks like tying knots, sealing packages, or unscrewing light bulbs.
The update also affects the ER 2 model, which DeepMind describes as a “high-level brain” for robots. The solution plans multi-step tasks, monitors whether actions are completed or need repetition, and allocates work among multiple robots.
In one video, Apollo 2 instructs a two-armed Google robot to organize tools into a container while cleaning a garage.
The company also showcased the local Gemini Robotics On-Device 2 model. It operates without internet connectivity and adapts to new robots with fewer than 200 examples in just a few hours of training.
Currently, Gemini Robotics 2 is available in private preview, while ER 2 is accessible via the Gemini API and Google AI Studio. On-Device 2 is being tested by trusted partners.
Earlier, in July, the company Humanoid introduced an approach for training robots on real-world production tasks.

[Оригинал](https://forklog.com/en/google-deepmind-unveils-enhanced-ai-model-for-robotics/)