Gemini Robotics 2.5 Achieves 96% Accuracy in Sorting Tasks During Mercedes-Benz Trials by Google DeepMind

Gemini Robotics 2.5 Excels in Industrial Testing

According to НВ — Техно: Google DeepMind has successfully demonstrated its Gemini Robotics 2.5 system on Apptronik Apollo 2 industrial robots. These trials took place on Mercedes-Benz production lines in Kecskemét, Hungary, and Sindelfingen, Germany, where the robots achieved an impressive 96% accuracy in sorting automotive components. This marks a significant milestone in applying advanced AI-driven robotics to large-scale manufacturing environments.

The Apollo 2 robots autonomously retrieved metal brackets from containers and organized them into trays according to the specific sequence required in vehicle assembly. Gemini Robotics 2.5 translates visual inputs and text instructions into precise physical actions, enabling the robots to handle delicate objects effectively. Google tested this system across more than 200 different tasks, including:

• feeding with tongs;
• tying ribbons on gift boxes;
• pipetting liquids.

Compared to previous models, Gemini Robotics 2.5 demonstrated a 30 to 100% improvement in task success rates and completed tasks 20 to 30% faster. Safety performance also saw notable gains, with successful obstacle navigation improving by 33% and safe human interaction increasing by 43%. To evaluate the system’s resistance to hazardous commands, an open benchmark called ASIMOV-Jailbreak was introduced, containing hundreds of scenarios testing the robot’s ability to identify and reject dangerous instructions.

Training with Gemini 4 Argon Simulation Technology

The Gemini 4 Argon platform supports robot training through thousands of simulated exercises based on real-world video data. By creating digital replicas of physical environments, it analyzes failed attempts and suggests improvements. The 96% success rate reflects results from demonstration trials rather than full-scale deployment. Google plans to gradually roll out Gemini Robotics 2.5 capabilities to its robotics partners.

"The successful testing of Gemini Robotics 2.5 represents a major step forward in automating manufacturing processes." - Google DeepMind

The combination of enhanced accuracy and improved safety paves the way for broader adoption of robotic automation across industries. As AI technology continues to evolve, integrating such systems could revolutionize production workflows and boost operational efficiency. Moreover, expanding access to this technology among partners may accelerate its global dissemination and impact.

In addition to advancements in robotics, Google has also unveiled new capabilities with its Gemini 4 Carbon platform, which focuses on enhancing security through vulnerability detection and secure code generation. This parallel development underscores Google's commitment to integrating AI across various sectors, further showcasing the potential of intelligent systems in both manufacturing and cybersecurity.