Google DeepMind has released Gemini Robotics ER 1.6, an updated version of its embodied reasoning system designed to improve autonomous robot performance in real-world tasks. The model focuses on spatial reasoning and multi-view scene understanding, addressing key challenges in robot navigation and manipulation. According to the company’s blog post, ER 1.6 enhances a robot’s ability to interpret its surroundings by combining data from multiple camera angles and sensors. This update follows the initial launch of the Gemini Robotics framework in late 2023, which introduced foundational capabilities for robots to reason about their environment.
The new version introduces improvements in 3D scene reconstruction, allowing robots to build more accurate maps of their surroundings. This is particularly useful in warehouse settings, where robots must navigate dynamic environments filled with moving objects and changing layouts. The system also strengthens multi-view consistency, ensuring that a robot’s understanding of an object remains stable even when viewed from different angles. These advancements are part of Google DeepMind’s broader effort to bridge the gap between simulation and real-world robotics deployment.
Early tests show ER 1.6 reduces errors in object recognition by 12% compared to its predecessor. The update also improves performance in tasks requiring fine motor control, such as picking up irregularly shaped items. Google DeepMind states the model has been tested on physical robots, including those operating in logistics centers, though specific deployment details remain limited.
The release comes as competition in the robotics field intensifies, with companies like NVIDIA and Tesla also advancing their own embodied AI systems. Google DeepMind has not announced commercial partnerships or pricing for ER 1.6, but the technology is expected to influence future robotics applications in manufacturing and service industries.
Source: deepmind.google