Gemini Robotics: AI meets the real world
With the Gemini Robotics and Gemini Robotics-ER AI models based on Gemini 2.0, Google DeepMind aims to drive robotics forward: The basic model masters physical tasks, while ER solves complex problems in dynamic environments.
Google DeepMind wants to create robots that act and think independently. On the way there, the company has now presented a new milestone in the combination of artificial intelligence (AI) and robotics : the Gemini Robotics and Gemini Robotics-ER (Embodied Reasoning) models based on Gemini 2.0. Both models aim to make AI systems capable of acting not only in the digital world, but also in the physical world.
While Gemini Robotics focuses on the basics of physical interaction, Gemini Robotics-ER complements these capabilities with logical reasoning for complex problems. The focus here is on the development of robots that can perform tasks independently in dynamic environments - from warehouse logistics to everyday assistance.
Three core innovations are driving development
Gemini Robotics' progress is based on three technological pillars:
1. universality
The ability to apply AI models universally to different robots and tasks - without customisation. For example, the same AI can control both a robotic arm in manufacturing and a mobile robot in logistics. The AI can also deal with and master situations that have never been covered in training.
2. interactivity
Like Gemini 2.0, Gemini Robotics is intuitive and interactive. The AI understands different languages as well as everyday language and can respond to complex instructions. As it continuously monitors the environment, it recognises changes and can react to them dynamically.
3. dexterity
The system helps with the precise physical manipulation of objects in 3D spaces. It can grasp fragile objects, fold paper or stack boxes without prior programming.
Enhanced capabilities through embodied reasoning
Gemini Robotics-ER builds on the three foundations and integrates the ability for better logical reasoning in real time. This is intended to improve spatial thinking in particular. The model is designed to enable robots to solve complex problems in unpredictable environments - for example by planning chains of action, setting priorities or recognising cause-and-effect relationships. Thanks to Gemini, robots can also learn completely new skills through spatial thinking in combination with programming capability - in other words, they can act intuitively.
Source: Google DeepMind
If a solution to a problem cannot be found on its own, the model can also follow a human demonstration in order to learn contextually.
In the long term, the new AI models are to be used in industry, disaster relief and as everyday assistance. Gemini Robotics is intended to automate repetitive physical tasks, while Gemini Robotics-ER acts as a problem solver in unpredictable contexts.
14 people like this article
I find my muse in everything. When I don’t, I draw inspiration from daydreaming. After all, if you dream, you don’t sleep through life.