Google DeepMind Releases Gemini Robotics-ER 1.6 with Enhanced Robot Reasoning
The new model can read analog gauges and call Google Search to understand its physical surroundings.
Google DeepMind has launched Gemini Robotics-ER 1.6, a significant upgrade to its AI model for robotic systems. The model is engineered with enhanced visual and spatial awareness, allowing robots to better infer their surroundings. A standout new capability is 'Agentic Vision,' which enables robots to read analog instruments by magnifying images to estimate proportions and intervals, a feature developed in direct response to partner Boston Dynamics' needs. Furthermore, the model can natively call external tools like Google Search to retrieve information, combining visual, linguistic, and behavioral models to plan and execute more useful physical tasks.
Beyond perception, Gemini Robotics-ER 1.6 demonstrates major improvements in safety and multi-view understanding. It now adheres to specific physical safety constraints, such as 'do not handle liquids' or 'do not lift objects weighing more than 20 kg,' and shows better hazard identification. The enhanced multi-view inference function allows robots to properly understand relationships between images from multiple cameras, crucial for navigation in complex environments. According to Google, this leap in embodied reasoning is essential for bridging the gap between digital intelligence and practical, real-world utility in daily life and industry.
- Introduces 'Agentic Vision' for reading analog instruments, developed with Boston Dynamics, by magnifying images for accurate measurements.
- Can natively call tools like Google Search and user-defined functions to retrieve information and reason about tasks.
- Shows significantly improved safety, adhering to constraints like not lifting >20kg objects and better identifying surrounding hazards.
Why It Matters
Moves robots from simple instruction-following to true physical-world reasoning, enabling safer, more autonomous deployment in complex industrial and daily tasks.