Google DeepMind has taken a significant leap forward in robotics, unveiling two new AI models—Gemini Robotics 1.5 and Gemini Robotics-ER 1.5—that give robots the ability to reason, plan, and even use the internet to solve complex tasks. This marks a major step toward creating versatile, adaptable robots that don’t require extensive reprogramming for every new scenario.
The breakthrough is centered on Gemini Robotics-ER 1.5, where “ER” stands for “embodied reasoning.” This system allows a robot to think through a problem and plan its actions before making a move, much like a text-based chatbot reasons through a query. In demonstrations, a robot was asked to sort recyclables “according to local regulations.” The machine identified its location, searched online for the specific local rules, and then correctly sorted the items. In another impressive test, a robot tasked with packing for a London trip first checked the weather forecast online and then packed appropriate clothing.
This new approach uses a two-model system: Gemini Robotics-ER 1.5 does the thinking and planning, while Gemini Robotics 1.5 executes the physical actions. One of the most powerful aspects of this technology is its cross-platform compatibility. A task learned by one type of robot, like a two-armed research model, can be seamlessly transferred to completely different hardware, including a manufacturing arm or even a full-scale humanoid robot like Apptronik’s Apollo. This could dramatically accelerate robotics development across all industries.
Google DeepMind is making the “thinking” model, Gemini Robotics-ER 1.5, available to developers through an API, while the action model remains limited to select partners for now. This announcement demonstrates a clear push towards creating truly intelligent robots that can navigate and interact with the complexities of the real world.