.svg){kind=spacer}
Details
- On September 26, 2025, Google DeepMind unveiled Gemini Robotics 1.5, a suite of vision-language-action models designed specifically for real-world robots.
- This release adapts the multimodal Gemini 1.5 transformer with specialized motor-control modules and leverages a massive dataset of 10 million trajectories across household, industrial, and mobile manipulation tasks.
- Demonstrations include a robotic arm assembling furniture, sorting groceries, and loading dishwashers—all without rigidly programmed paths.
- The system uses RT-X training methods, honing skills first in simulation and then efficiently transferring them to hardware, reducing costly real-world testing by half compared to previous solutions.
- Google Cloud Robotics now offers an API for enterprises to describe tasks in natural language and generate instantly executable motion plans.
- Safety is prioritized with built-in Reflex guardrails, pausing operation if sensors or computer vision detect anomalies.
- Software development kits for ROS 2 and Nvidia Isaac will roll out to select partners later in 2025, with broader availability planned for early 2026.
Impact
This launch escalates the rivalry with players like Tesla Optimus and Figure 01, who still depend on custom task pipelines. By turning advanced robotics AI into a cloud-based service, Google DeepMind lowers adoption barriers for factories and logistics companies. The built-in Reflex safety features help the technology align with new EU regulations, potentially accelerating widespread European deployment.
