Google DeepMind has introduced Gemini Robotics, a breakthrough in AI-driven robotics, marking a significant step toward bringing artificial intelligence into the physical world. Building on the capabilities of Gemini 2.0, these new models — Gemini Robotics and Gemini Robotics-ER — expand AI’s potential beyond digital interactions by enabling real-world reasoning and action.
Gemini Robotics is an advanced vision-language-action (VLA) model designed to incorporate physical actions as an output modality, allowing AI to directly control robotic systems. Meanwhile, Gemini Robotics-ER enhances spatial reasoning, giving roboticists the ability to integrate Gemini’s embodied reasoning (ER) into their own systems.
According to a Google DeepMind blog post, these models allow robots to perform a wider range of real-world tasks with increased adaptability and precision. Gemini Robotics can generalize across novel situations, respond to conversational commands in multiple languages, and execute dexterous tasks requiring fine motor skills. DeepMind researchers reported that the model more than doubles performance benchmarks compared to previous state-of-the-art VLA models.
To push AI-powered robotics forward, Google DeepMind is collaborating with Apptronik to develop next-generation humanoid robots. Additionally, Gemini Robotics-ER is being tested by select industry partners, including Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools.
As part of its commitment to responsible AI development, Google DeepMind is integrating safety measures that ensure robotic systems operate within predefined ethical and physical parameters. The company is also launching a new dataset, ASIMOV, to evaluate and enhance safety in embodied AI.
With Gemini Robotics, Google DeepMind is redefining the future of AI-powered robotics, unlocking new possibilities for intelligent, adaptable, and safe automation.
