Google Launches Gemini Robotics 2.5 for Humanoid Robots
Google DeepMind's Gemini Robotics 2.5 helps robots handle delicate objects and factory work, with a 96% success rate in car-plant tests of the Apollo 2 humanoid.

Google has launched Gemini Robotics 2.5, a new AI model designed to let robots carry out more complex and precise tasks, from handling fragile objects to repetitive factory work. Carolina Parada, vice president at Google DeepMind and head of its robotics division, announced the release and called it the company's most advanced model combining vision, language and action. Such models let a robot understand its surroundings, follow instructions and act on them.
Tested in car factories
Google tested the model with Apptronik and Mercedes-Benz, using the Apollo 2 humanoid robot in car plants in Hungary and Germany. In those trials, the robots picked car parts out of bins holding large quantities of components and arranged them in trays in the required order. The success rate reached 96%, helping automate repetitive work that previously needed technicians.
More broadly, the model was evaluated on more than 200 varied tasks, including serving food with tongs, tying a gift ribbon and moving liquids with precision tools. According to Google, success rates rose by between 30% and 100% compared with earlier models, depending on the task, and tasks are completed 20% to 30% faster.
Safety measures
Google also focused on safety. The model can recognise hazards and nearby people and adjust its movements in real time. Evaluations showed a 33% improvement in completing tasks safely when obstacles are present, and 43% when people are close to the robot.
The company also unveiled ASIMOV-Jailbreak, an open-source benchmark that tests whether robots can resist unsafe instructions disguised as ordinary requests, helping gauge how well they keep to safety rules in unpredictable settings.
The release points to Google's push to move humanoid robots beyond repetitive motions on production lines toward handling delicate objects, understanding instructions and adapting to change, which could widen their use in factories and other workplaces.
Apps worth a look
Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.
Talk to an AI conversation partner; every sentence comes back corrected and explained in your language. 9 languages, A1 to C2.
Open EDIFACT and X12 messages, see every segment explained, convert to JSON/XML, export PDF and Excel. Runs locally, no upload.
Pick who it is for, the occasion and their personality. The search engine reads product descriptions from local shops and ranks what fits. Free, no sign-up.
More from this day
See all →
Unacademy CEO Gaurav Munjal Steps Down; Akhilendra Kumar Takes Over

India Bans Export of Lithium-Ion Battery Black Mass

Greenjoules Closes ₹150 Cr Series B Led by Amazon's Smbhav Fund



