WORLDTECHGoogle DeepMind releases Gemini Robotics 2.5 with stronger dexterity for industrial robots
Google DeepMind has introduced Gemini Robotics 2.5, an updated vision-language-action model intended to help robots handle demanding physical jobs in factories and everyday environments. The company says the release broadens the range of tasks robots can take on while raising their dexterity, reliability and safety.
Results across more than 200 tasks
The model was put through more than 200 tasks, including handling fragile items, working with tools, manipulating pliable materials and carrying out precision assembly.
According to Google DeepMind, Gemini Robotics 2.5 posted success rates between 30 and 100 percent higher than previous versions, depending on the task, and finished operations 20 to 30 percent faster.
Humanoid trial on Mercedes-Benz lines
The most notable demonstration involved Apptronik's Apollo 2, a humanoid robot roughly shaped like a person, working on Mercedes-Benz production lines in Kecskemét, Hungary, and Sindelfingen, Germany.
There, the robots carried out automotive kitting: picking individual parts out of bulk bins and placing them into trays for vehicle assembly. Apollo 2 drove its arms and hands with Gemini Robotics 2.5 and planned its work with Gemini Robotics ER, reaching a reported 96 percent success rate in the proof-of-concept programme.
Open questions
- Will the reported 96 percent success rate hold beyond the proof-of-concept programme?