Technology and Scienceupdated 12m ago2 min read
DeepMind Unveils Gemini Robotics 2.5 With Higher Success Rates for Factory Humanoids
8outlets comparedWhere the coverage agrees
3 Other2 Western Alternative1 Local Western2 Other source typesThis article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.
Apollo 2 humanoid achieved 96% success sorting automotive parts in Mercedes-Benz plants. Gemini Robotics 2.5 delivers 30–100% task improvements and 20–30% faster completion.
Save coverage image
8 outlets in this story. Read their headlines below.
Gemini Robotics 2.5 Debut
Google DeepMind revealed Gemini Robotics 2.5 as an update to its AI-powered robotics platform, pairing Gemini Embodied Reasoning with Gemini Robotics to translate instructions and live video into robot actions. DeepMind said Gemini Robotics 2.5 improved performance on tasks requiring high dexterity, and it evaluated the system across more than 200 activities including serving food with tongs and using a pipette.
DeepMind also showed Apptronik’s Apollo 2 humanoid robot tying a bow with its five-fingered hands, and DeepMind said the release achieved a 30 to 100 percent higher success rate and was 20 to 30 percent faster than the last version in laboratory testing. DeepMind reported that Gemini-powered Apollo 2 robots kitted automotive parts in a few European factories with a 96 percent success rate, and DeepMind director Kanishka Rao said closing the remaining gap would be a “steep road.”
Factory Pilots and Partners
Google DeepMind partnered with Apptronik and Mercedes-Benz to test Gemini Robotics 2.5 on active production lines, and the company reported that Apollo 2 sorted and placed parts in the exact order each vehicle required. IoT News said Apollo 2 performed automotive kitting at plants in Kecskemét, Hungary, and Sindelfingen, Germany, with a 96 percent success rate during a proof-of-concept on live production lines.
Apptronik said in a statement that “This is what we built Apollo for: useful work alongside people in real industrial settings,” and Apptronik added that pairing Apollo 2 with Gemini Robotics improved results with “very little drop-off” on Mercedes-Benz production. DeepMind also described a two-model setup for the pilot, with Gemini Robotics ER tracking the workflow while Gemini Robotics 2.5 controlled the robot’s hands and arms.
Open Safety Benchmark ASIMOV
DeepMind introduced the open safety benchmark ASIMOV-Jailbreak to test whether robotic AI can resist disguised unsafe requests, and the benchmark is described as being released on Hugging Face with 1,500 prompts. IoT News said Argon achieved a 99.5 percent jailbreak resistance rate on the benchmark, and it contrasted that with an 82 percent figure for Gemini Robotics ER 2.
Mezha.net reported that DeepMind also introduced an open database of 1,500 prompts for testing robotic AI safety, and it said the system’s safety level reached 98.8 percent after adding Argon. DeepMind said Gemini Robotics 2.5 uses inference-time reasoning guidance to steer robots around physical hazards, and it described safe task completion improving by 33 percent near obstacles and by 43 percent when people were in the way.