On July 30, 2026, Google DeepMind announced the latest AI model for humanoid robots, Gemini Robotics 2. This model advances from upper-body control to full-body control, enabling walking, bending, and advanced dexterity, as well as collaborative tasks among multiple robots.
Gemini Robotics 2 translates visual and language inputs into motion control, supporting full-body and dual-arm robot operations. It features enhanced physical reasoning capabilities with Gemini Robotics ER 2, which manages human communication and multi-step task planning, and Gemini Robotics On-Device 2, which adapts quickly to new robots with minimal data.
The model demonstrated its capabilities using Apptronik's humanoid robot Apollo 2, performing tasks such as walking to a table, lifting objects, and accurately storing them. With a 22-degree-of-freedom hand, it can execute delicate tasks like tying knots. The introduction of the ASIMOV-Agentic benchmark emphasizes safety, allowing the robot to detect human presence and stop safely, marking it as the safest model to date.
Editor's Note
The introduction of Gemini Robotics 2 signifies a pivotal moment in humanoid robotics, enhancing both physical capabilities and safety measures. As robots become more adept at complex tasks and human interaction, enterprises must consider the implications for workforce integration and operational efficiency. The focus on safety with the ASIMOV-Agentic benchmark reflects an industry-wide commitment to responsible AI deployment.
Leave a comment