The development

What happened

Google DeepMind has introduced Gemini Robotics 2.5, a vision-language-action AI model that turns visual observations and instructions into robot actions. Google says it is rolling out to robotics partners and reports a Mercedes-Benz automotive-parts kitting pilot using Apptronik's Apollo 2, not general availability or proof of universal reliability.

Key facts

  • Google says it evaluated the model across more than 200 tasks, including tool use and handling deformable or delicate objects. Those are publisher-run evaluations rather than independent measures of every deployment.

  • The reported proof-of-concept operated on production lines in Kecskemét, Hungary, and Sindelfingen, Germany. Apollo 2 sorted automotive parts from bins into sequencing containers.

  • Gemini Robotics ER tracked the workflow while Robotics 2.5 controlled hands and arms. Google reports 96% success for this kitting pilot; the figure is not a reliability guarantee for other tasks or settings.

  • The announcement also describes inference-time guidance around physical hazards and a separate ASIMOV-Jailbreak benchmark. Those safety evaluations should not be conflated with the factory-pilot success rate.

The context

Why it matters

The pilot moves a learned robot-action model into an active industrial setting while retaining a limited proof-of-concept scope. TideReach assessment: broader adoption needs task-specific validation, failure handling and physical-safety assessment; partner rollout and a reported success rate do not establish general-purpose deployment or certification.