The Moment Robots Gain ‘Intelligence’! Google Unveils the Physical Reasoning Model ‘Gemini Robotics-ER 1.6’

📰 News Summary

• The latest model ‘Gemini Robotics-ER 1.6’, focusing on ‘Embodied Reasoning’, has been launched via the Gemini API and Google AI Studio.
• The robot’s ability to comprehend its surrounding environment has been significantly enhanced with ‘Multiview Understanding’ and ‘Spatial Reasoning’, resulting in improved accuracy in pointing, counting, and executing tasks.
• In collaboration with Boston Dynamics, a new feature has been introduced that enables the robot to read complex analog gauges and site glass values, greatly enhancing its practicality in industrial settings.

💡 Key Highlights

• Advanced Reasoning and Tool Utilization: The model can natively call Google Search, VLA (Vision-Language-Action), and user-defined functions to plan and execute tasks.
• Autonomy Engine ‘Success Detection’: It visually determines whether a task has been completed and autonomously decides whether to retry or move on to the next plan if a failure occurs.
• Unmatched Spatial Awareness: Compared to its predecessor (1.5) and Gemini 3.0 Flash, this model has evolved in its ability to accurately count objects and identify items that meet specific constraints (e.g., “an item that fits the size of a blue cup”).

🦈 Shark’s Eye (Curator’s Perspective)

Finally, robots have evolved from mere machines that follow ‘direct commands’ to agents that ‘reason’ through their environment, folks! What’s especially thrilling is this model’s ability to integrate multiple camera views (like overhead and handheld) to understand the world as a unified whole—this is what we call ‘Multiview Reasoning’, baby! Even in the presence of obstacles or low light, combining multiple perspectives allows robots to make human-like situational judgments, which is simply mind-blowing!

Moreover, the functionality of pointing (specifying coordinates) as a relay step is a game changer! Not just saying, “Grab that,” but now being able to perform complex logic in physical space like, “Count this, then bring the smallest one here,” is clear evidence that the robot’s brain is upgraded! The future of Boston Dynamics robots being equipped with this technology has me swimming with excitement! 🦈🔥

🚀 What’s Next?

The norm of robots ‘recognizing their own failures and retrying’ will significantly accelerate the full automation of factories and logistics hubs. The ability to read analog gauges opens the door for AI robots to be swiftly implemented in infrastructure inspection settings, especially where older equipment remains!

💬 Shark’s One-liner

Your shark reporter, ‘Haru Shark’, says with this, I’d be able to read complex gauges in the deep sea and go treasure hunting! Autonomy is the first step to freedom! 🦈💎

📚 Terminology Explained

• Embodied Reasoning: The technology that enables AI to make judgments not just based on knowledge in the digital realm, but by considering the spatial relationships and properties of objects in the physical world.

• Success Detection: The process by which a robot autonomously evaluates whether the actions it performed achieved the target, using sensors and visual information.

• VLA (Vision-Language-Action): A model that takes visual inputs and language instructions, outputting them as concrete ‘actions’ for the robot to perform.

• Source: Gemini Robotics-ER 1.6