Feed
Google DeepMind Launches Gemini Robotics 2.5 Using Simulated Training Data
AI

Google DeepMind Launches Gemini Robotics 2.5 Using Simulated Training Data

10 Nov 2026, 9:38 pm · 10d ago · 1 min read · OfficeChai

Google DeepMind has launched Gemini Robotics 2.5, its most advanced vision-language-action model designed for complex robotic control and task execution. To accelerate the model's physical reasoning capabilities, Google deployed Gemini 4 Argon as an automated simulation engine to generate synthetic training datasets at scale. This synthetic data generation method bypassed the traditional bottlenecks of real-world robotic data collection. DeepMind’s dual-model framework demonstrates how frontier multimodal models can train and bootstrap specialized physical AI agents, marking a critical step forward in bridging generative AI software with autonomous hardware robotics across diverse industrial and consumer environments.